Model family by OpenAI

gpt-oss

OpenAI's open-weight GPT models under the Apache 2.0 license.

Where it sits in OpenAI's lineup

GPT TerraOpenAI's fast, efficient GPT tier for visual reasoning, computer use, and writing.GPT-6.1 Terra
GPT SolOpenAI's GPT model tier for demanding everyday and agentic work.GPT-6.1 Sol
GPT LunaOpenAI's most efficient GPT model tier for focused, high-volume tasks.GPT-6 Luna
GPT AstraOpenAI's highest-capability GPT model tier.GPT-6 Astra
GPT-5OpenAI's GPT-5 frontier models, released before the Sol, Terra and Luna tiers.GPT-5.5
gpt-ossOpenAI's open-weight GPT models under the Apache 2.0 license.This model
GPT CodexOpenAI's GPT models tuned for agentic coding in Codex.GPT-5.3-Codex

Made by: OpenAI
Releases: 1 tracked
Current release: gpt-oss-120b

Strengths and weaknesses

Compare with other models Up to 3 beside gpt-oss-120b

gpt-oss-120b

Suggested comparisons

The models a buyer weighs against gpt-oss-120b, picked on list price and on Artificial Analysis's Intelligence Index of overall quality. Add one to draw it beside gpt-oss-120b.

Search models by name, maker or family

gpt-oss-120bOpenAIThis page, always on the chart

Among the 40 current releases on this chart, gpt-oss-120b's best rank is 24th of 25 on Answers in full. Agentic coding is Terminal-Bench's run on Terminal-Bench 2.0, Math is Epoch AI's run on OTIS Mock AIME 2024-2025 and Factuality is OpenAI's own figure on SimpleQA, marked on the chart. It ranks lowest on Instruction following, 8th of 8 and Science, 24th of 24. It cannot take Computer use and Multimodal. gpt-oss-120b reads text only.

Each axis is one public benchmark at one setting, on its own fixed 0 to 100 scale, so adding a model never moves another. A dotted bridge crosses a missing result, which is never a zero. A hollow mark is a result from another source: the maker's own figure or another lab's run, where the benchmark's usual source has not tested the model yet. The readout names it.

On the chart

gpt-oss-120bThis pageResults on 10 of 12 axes$0.09 / $0.36On Google Cloud Vertex / Agent Platform (Google managed); the maker sells no API access236 tokens/sFirst token in 0.8 s

Prices and speeds as of Oct 28, 2026.

gpt-oss-120b on each axis10 of 12 with a result

The share of Terminal-Bench 4.0's long terminal tasks the model finishes on its own: services, kernels, proofs and data pipelines, graded on the final artifact with no partial credit.
Terminal-Bench 4.0, pass@1 averaged over 3 runs, mini-swe-agent harness. Where it has no run of a model, the maker's own figure or another lab's run stands in, marked with its benchmark.
- gpt-oss-120b14.2%Terminal-Bench's runBenchmark: Terminal-Bench 2.0, accuracy, Terminal-Bench 2.0 leaderboard, mini-swe-agent harness. From another source, so it is drawn and marked but not ranked with the others.The Terminal-Bench team's run on its version 2.0 leaderboard with the mini-swe-agent harness, the same agent the axis uses. Terminal-Bench 2.0 is the earlier, easier version of the test.
The share of SciCode's scientific programming subproblems the model's code solves on the first attempt.
SciCode, subproblem accuracy, pass@1, highest effort run
- gpt-oss-120b34%
The share of FrontierMath's held-out research-grade problems (tiers 1 to 3) the model answers correctly.
FrontierMath Tiers 1-3, accuracy, pass@1, highest effort run. Where it has no run of a model, the maker's own figure or another lab's run stands in, marked with its benchmark.
- gpt-oss-120b88.9%Epoch AI's runBenchmark: OTIS Mock AIME 2024-2025, accuracy, high reasoning. From another source, so it is drawn and marked but not ranked with the others.Epoch AI's own run at high reasoning, the highest gpt-oss-120b offers. Epoch has not run it on FrontierMath; this competition-math set is easier.
The share of GPQA Diamond's graduate-level biology, chemistry and physics questions the model answers correctly.
GPQA Diamond, accuracy, pass@1, highest effort run
- gpt-oss-120b75.8%
The share of Humanity's Last Exam's 2,158 text-only questions, expert-written at the edge of human knowledge, the model answers correctly without tools.
Humanity's Last Exam, accuracy, text-only questions, no tools
- gpt-oss-120b19.6%
The share of each AutomationBench task's objectives the model completes by calling REST APIs across simulated business apps (Gmail, Sheets, Slack, Salesforce, Zendesk, Jira, HubSpot), averaged over tasks. A task that breaks one of its policy guardrails scores zero.
AutomationBench-AA, score, Artificial Analysis run
- gpt-oss-120b0.2%Artificial Analysis publishes 0.2 for its high-effort run on this test, far below the model's other results.
How much of each OSWorld 2.1 workflow the model completes by operating a real Ubuntu desktop through screenshots, mouse and keyboard: all 108 long tasks across apps, files, mail and browsers, on the September 2026 v2.1 release of the task files. A point is the mean share of each task's weighted checks the model passes (the partial score, as a percentage). The OSWorld team's own runs come first. Where the team has not run a release, the point is another lab's run or the maker's own published figure, marked with who ran it and the benchmark it ran: OSWorld 2.1 in the maker's harness, OSWorld 2.0's offline set or the earlier OSWorld-Verified. Those points are drawn so the release has a figure, and they are not ranked against the team's runs.
OSWorld 2.1, partial score, full 108-task set, v2.1 release
- gpt-oss-120bgpt-oss-120b reads text only.
The share of AA-LCR's questions the model answers correctly by reasoning across long sets of real documents, about 100K tokens each.
AA-LCR v1.1, accuracy, Artificial Analysis run
- gpt-oss-120b52%
The share of MMMU-Pro's college-level questions over charts, diagrams and photos the model answers correctly.
MMMU-Pro, overall accuracy, no tools
- gpt-oss-120bgpt-oss-120b reads text only.
The share of AA-Omniscience's 6,000 knowledge questions across 42 economically relevant topics the model answers correctly from memory, without search.
AA-Omniscience Accuracy, accuracy, Artificial Analysis run. Where it has no run of a model, the maker's own figure or another lab's run stands in, marked with its benchmark.
- gpt-oss-120b16.8%reported by OpenAIBenchmark: SimpleQA, accuracy, no browsing. From another source, so it is drawn and marked but not ranked with the others.OpenAI's own run in its model card, without browsing. It answered 78.2% of questions wrongly; OpenAI counts that share of all questions as its hallucination rate.Hallucination rate (wrong answers as a share of all questions, OpenAI's definition) 78.2% in the same run (lower is better)
The share of SpeechMap's 2,120 sensitive and controversial but legal requests the model answers in full: arguments, essays and explanations on contested themes in politics, religion, race, sex, history and similar topics, four wordings per theme across 530 themes. SpeechMap's judge rates each response complete, evasive (hedged or redirected), a denial, or an error (the provider's API blocked the request). Only complete answers count, so the point is the share answered in full: 100 minus the share hedged, refused or blocked by the provider. Higher means more lawful requests answered in full. The share blocked at the provider is printed apart in the table, because an API error is not the model declining. Where SpeechMap runs a release both with reasoning off and with reasoning on (a -reasoning or -thinking entry), the reasoning-on entry is used; a release with one entry uses that entry.
SpeechMap, % Complete over 2,120 contested requests, reasoning-on entry
- gpt-oss-120b36.3%SpeechMap counted 63.3% of requests denied and 0.3% evaded.
The share of MultiChallenge's multi-turn conversations in which the model keeps every instruction it was given earlier in the conversation.
MultiChallenge, accuracy, Scale AI run
- gpt-oss-120b45.3%On Oct 28, 2026, Scale AI ranked it 31st.Plus or minus 2.15 points (95% interval)

Beside the benchmarks

Cost, speed, capacity, quality and accuracy for gpt-oss-120b, with where each figure sits among the 40 models GILD tracks. These measures stand beside the chart and never feed it.

Cost

$0.11per task

Cost per task. One Intelligence Index task at list price, every token counted. Run at high effort.

Speed

250tokens a second

Output speed. How fast it writes once the answer starts. Run at high effort.

First token: 0.8 s
Whole answer: 11 s

Capacity

131Ktokens

Context window. The most one request can hold, as the maker states it.

Max output: 131K
Reads: Text

Quality

11.6of 100

Intelligence Index. One composite score across independent evaluations. Run at high effort.

GDPval knowledge work: 616 Elo
Terminal-Bench 4.0: 0.0%
Epoch Capabilities Index: 140

Hallucination

90.8%wrong when it should say “I don’t know”

Hallucination rate. Of the questions it can't get right, how often it answers anyway. Lower is better. Run at high effort.

Knowledge accuracy: 21.8%

Where you can get it

7of 19 routes

Routes offering it. Of the clouds, gateways and coding products GILD tracks, those offering it today.

Open weights: Yes
Licence: Apache 2.0

Every route and price

Figures from Artificial Analysis, openai, Epoch AI and GILD's route matrix, read Oct 26 to 29, 2026. Each measure's definition and source open under . Not yet published for gpt-oss-120b: Cached input.

Every value with its benchmark, run and source

gpt-oss-120b's result on each axis, with the benchmark, the setting and run, who reported it, the source and the date
Axis	Benchmark and setting	gpt-oss-120b	Run	Source	Date
Agentic coding	Terminal-Bench 2.0Accuracy, Terminal-Bench 2.0 leaderboard, mini-swe-agent harness. The Terminal-Bench team's run on its version 2.0 leaderboard with the mini-swe-agent harness, the same agent the axis uses. Terminal-Bench 2.0 is the earlier, easier version of the test.Stands in for Terminal-Bench 4.0, pass@1 averaged over 3 runs, mini-swe-agent harness, which has no run of gpt-oss-120b yet.	14.2%	Effort: not published. Harness: mini-swe-agent.Not drawn: Terminal-Bench (via Epoch AI) 18.7% in another lab's harness, Nov 1, 2025. Source	Terminal-Bench (via Epoch AI)Run by Terminal-Bench (via Epoch AI)	Nov 3, 2025
Code	SciCodeSubproblem accuracy, pass@1, highest effort run	34%	gpt-oss-120b (High)Effort: high. Harness: none.	Artificial Analysis	Oct 28, 2026
Math	OTIS Mock AIME 2024-2025Accuracy, OTIS Mock AIME 2024-2025, high reasoning. Epoch AI's own run at high reasoning, the highest gpt-oss-120b offers. Epoch has not run it on FrontierMath; this competition-math set is easier.Stands in for FrontierMath Tiers 1-3, accuracy, pass@1, highest effort run, which has no run of gpt-oss-120b yet.	88.9%	Effort: high. Harness: none.	Epoch AIRun by Epoch AI	Dec 11, 2025
Science	GPQA DiamondAccuracy, pass@1, highest effort run	75.8%	Effort: high. Harness: none.	Epoch AI	Dec 11, 2025
Hard reasoning	Humanity's Last ExamAccuracy, text-only questions, no tools	19.6%	gpt-oss-120b (High)Effort: high. Harness: none.	Artificial Analysis	Oct 28, 2026
Tool use	AutomationBench-AAScore, Artificial Analysis run. Artificial Analysis publishes 0.2 for its high-effort run on this test, far below the model's other results.	0.2%	gpt-oss-120b (High)Effort: high. Harness: Artificial Analysis AutomationBench harness.	Artificial Analysis	Oct 28, 2026
Computer use	OSWorld 2.1Partial score, full 108-task set, v2.1 release	gpt-oss-120b reads text only.
Long context	AA-LCR v1.1Accuracy, Artificial Analysis run	52%	gpt-oss-120b (High)Effort: high. Harness: none.	Artificial Analysis	Oct 28, 2026
Multimodal	MMMU-ProOverall accuracy, no tools	gpt-oss-120b reads text only.
Factuality	SimpleQAAccuracy, SimpleQA, no browsing. OpenAI's own run in its model card, without browsing. It answered 78.2% of questions wrongly; OpenAI counts that share of all questions as its hallucination rate.Stands in for AA-Omniscience Accuracy, accuracy, Artificial Analysis run, which has no run of gpt-oss-120b yet.	16.8%Hallucination rate (wrong answers as a share of all questions, OpenAI's definition) 78.2% in the same run (lower is better)	Effort: not published. Harness: none.	OpenAIReported by OpenAI	Aug 5, 2025
Answers in full	SpeechMap% Complete over 2,120 contested requests, reasoning-on entry. SpeechMap counted 63.3% of requests denied and 0.3% evaded.	36.3%	Effort: reasoning on. Harness: none.	SpeechMap.AI	Oct 28, 2026
Instruction following	MultiChallengeAccuracy, Scale AI run. On Oct 28, 2026, Scale AI ranked it 31st.	45.3%Plus or minus 2.15 points (95% interval)	Effort: not published. Harness: none.	Scale AI SEAL	Oct 28, 2026

Price and speed

OpenAI sells no API access to gpt-oss-120b. Among Amazon Bedrock, Google Vertex AI and Microsoft Foundry, the lowest list price is on Google Cloud Vertex / Agent Platform (Google managed): $0.09 per million input tokens and $0.36 per million output tokens. Every host's price is in Routes. Google Cloud pricing, Oct 29, 2026.
gpt-oss-120b, at high effort, writes 236.4 output tokens per second at the median across the providers serving it, and its first token arrives 0.8 s after the request. Artificial Analysis, Oct 28, 2026.
Fastest on the chart is the shortest time to a 500-token answer: time to first token plus 500 tokens at the median output speed.

Every model with a result, axis by axis

57 models. A maker's own figure and another lab's run are marked with their benchmark; a missing result reads "Not yet published", never zero.

The result of every model GILD has benchmark results for, on each of the 12 axes. A vendor-reported value is marked. A model's run notes open under its row from its name, each with the axes it covers.
Model	Agentic coding	Code	Math	Science	Hard reasoning	Tool use	Computer use	Long context	Multimodal	Factuality	Answers in full	Instructions
gpt-oss-120b	14.2%Run by Terminal-Bench (via Epoch AI)Terminal-Bench 2.0	34%	88.9%Run by Epoch AIOTIS Mock AIME 2024-2025	75.8%	19.6%	0.2%	Not yet published	52%	Not yet published	16.8%Reported by OpenAISimpleQA	36.3%	45.3%
Agentic coding. The Terminal-Bench team's run on its version 2.0 leaderboard with the mini-swe-agent harness, the same agent the axis uses. Terminal-Bench 2.0 is the earlier, easier version of the test. Terminal-Bench (via Epoch AI) Math. Epoch AI's own run at high reasoning, the highest gpt-oss-120b offers. Epoch has not run it on FrontierMath; this competition-math set is easier. Epoch AI Tool use. Artificial Analysis publishes 0.2 for its high-effort run on this test, far below the model's other results. Artificial Analysis Factuality. OpenAI's own run in its model card, without browsing. It answered 78.2% of questions wrongly; OpenAI counts that share of all questions as its hallucination rate. OpenAI Answers in full. SpeechMap counted 63.3% of requests denied and 0.3% evaded. SpeechMap.AI Instructions. On Oct 28, 2026, Scale AI ranked it 31st. Scale AI SEAL
GPT-6.1 Terra	49.5%	55.1%	90.9%	94.3%	46.2%	60.2%	74.2%Reported by OpenAIOSWorld 2.1	82.7%	84.7%	47.6%	Not yet published	74.2%
Computer use. OpenAI's own run at maximum reasoning effort, from its launch chart, at $1.19 per task. OpenAI does not say whether the score counts partial credit. OpenAI Instructions. On Oct 28, 2026, Scale AI ranked it 6th, tied with Muse Spark Open. Scale AI SEAL
GPT-6.1 Sol	55.1%	54.2%	93.7%	95.4%	52.9%	64.9%	78%Reported by OpenAIOSWorld 2.1	83%	86%	61.8%	69.5%	82%
Computer use. OpenAI's own run at maximum reasoning effort, from the GPT-6.1 Terra launch chart, at $1.15 per task against $8.17 for GPT-6 Astra. OpenAI does not say whether it counts partial credit. OpenAI Answers in full. SpeechMap tested it through OpenRouter at medium reasoning and counted 21.7% of requests denied and 8.8% evaded. SpeechMap.AI Instructions. On Oct 28, 2026, Scale AI ranked it 1st. Scale AI SEAL
GPT-6 Luna	13.6%	54.6%	78.9%	90.5%	38.5%	53.2%	48.9%Run by AnthropicOSWorld 2.1	83.3%	79.7%	39.6%	63.4%	Not yet published
Computer use. Anthropic's run of GPT-6 Luna through OpenAI's API with OpenAI's context compaction, 500 steps, averaged over five attempts, from the Claude Haiku 5.5 system card. Anthropic Answers in full. SpeechMap tested it through OpenRouter at medium reasoning and counted 27.4% of requests denied and 9.1% evaded. SpeechMap.AI
GPT-6 Astra	59.6%	56.5%	93.7%	95.8%	54.7%	68.5%	75.8%Run by Meta	80.7%	86.9%	61.6%	68.2%	Not yet published
Agentic coding. Vals.ai ranked it 4th on its leaderboard of Oct 24, 2026. Vals.ai Computer use. Meta's run of GPT-6 Astra at max effort in its own test framework on all 108 tasks: the average partial score. It fully completed 34.3% of tasks. Meta Answers in full. Reasoning cannot be turned off on this model. SpeechMap counted 21.7% of requests denied and 10.0% evaded. SpeechMap.AI
GPT-5.6 Sol	37.9%	57.1%	89.1%	93.5%	49.5%	60.1%	62.7%Run by OSWorld teamOSWorld 2.0	84%	83.4%	69.7%Run by Epoch AISimpleQA Verified	Not yet published	72.7%Run by Artificial AnalysisIFBench
Computer use. OSWorld team's official run of GPT-5.6 Sol at max reasoning, batch tool, 500 steps. OSWorld team Factuality. Epoch AI's run at max effort. SimpleQA Verified prints no hallucination rate, so none is shown beside it. Epoch AI Instructions. Artificial Analysis ran GPT-5.6 Sol at its highest published effort in its own evaluation harness. Artificial Analysis
GPT-5.6 Terra	22.7%	55%	86%	93.3%	42.9%	59.6%	Not yet published	83%	80.7%	42.3%	Not yet published	71.2%Run by Artificial AnalysisIFBench
Instructions. Artificial Analysis ran GPT-5.6 Terra at its highest published effort in its own evaluation harness. Artificial Analysis
GPT-5.6 Luna	11.6%	53.6%	82.1%	91.6%	39.5%	50.2%	Not yet published	83.7%	78.6%	41%Run by Epoch AISimpleQA Verified	Not yet published	Not yet published
Factuality. Epoch AI's run at max effort. SimpleQA Verified prints no hallucination rate, so none is shown beside it. Epoch AI
GPT-5.5	14.6%Run by Artificial AnalysisTerminal-Bench 4.0	55.8%	85.3%	93.5%Run by Artificial Analysis	45.8%	47.3%	49.5%Run by OSWorld teamOSWorld 2.0	84.3%	79.9%	63%Run by Epoch AISimpleQA Verified	Not yet published	75.9%Run by Artificial AnalysisIFBench
Agentic coding, Science and Instructions. Artificial Analysis ran GPT-5.5 at its highest published effort in its own evaluation harness. Artificial Analysis Computer use. The OSWorld team's own run of GPT-5.5 at xhigh reasoning with batched tool calls and a 500-step budget, on the first OSWorld 2.0 task release. OSWorld team Factuality. Epoch AI's run at xhigh effort, the highest it ran. SimpleQA Verified prints no hallucination rate, so none is shown beside it. Epoch AI
GPT-5.4	Not yet published	Not yet published	78.6%	93.3%	43.7%	Not yet published	Not yet published	82%	78.4%	Not yet published	Not yet published	73.9%Run by Artificial AnalysisIFBench
Instructions. Artificial Analysis ran GPT-5.4 at its highest published effort in its own evaluation harness. Artificial Analysis
GPT-6 Sol	44.4%	57.6%	89.8%	94.3%	47.9%	61.6%	64.4%Reported by OpenAIOSWorld 2.0	83.7%	82.9%	54.5%	64%	Not yet published
Computer use. OpenAI's own run at maximum reasoning effort, from its launch chart, at $3.25 per task. OpenAI Answers in full. SpeechMap tested it through OpenRouter at medium reasoning and counted 22.7% of requests denied and 13.2% evaded. SpeechMap.AI
GPT-5.3-Codex	Not yet published	Not yet published	Not yet published	91.5%Run by Artificial Analysis	42.5%	Not yet published	Not yet published	83.3%	78.5%	Not yet published	Not yet published	75.4%Run by Artificial AnalysisIFBench
Science. Artificial Analysis ran GPT-5.3-Codex at its highest published effort in its own evaluation harness; Epoch AI has no qualifying run. Artificial Analysis Instructions. Artificial Analysis ran GPT-5.3 Codex at its highest published effort in its own evaluation harness. Artificial Analysis
Claude Haiku 5.5	36.4%	55%	75.1%	89.6%	44.4%	35.4%	72.4%Reported by AnthropicOSWorld 2.1	82.7%	Not yet published	34.7%	78.8%	67.7%
Computer use. Anthropic's own run at max effort in its computer-use harness, averaged over five attempts, on the 82 tasks that need no internet. It fully completed 37.1% of them. Anthropic Answers in full. SpeechMap tested it at medium thinking and counted 18.3% of requests denied and 2.5% evaded. SpeechMap.AI Instructions. On Oct 28, 2026, Scale AI ranked it 10th. Scale does not say which reasoning setting it tested. Scale AI SEAL
Claude Sonnet 5.5	64.1%	61%	88.8%	95.6%	55%	71.8%	80.1%Reported by Anthropic	82.7%	Not yet published	51.5%	81.6%	Not yet published
Code, Hard reasoning, Tool use, Long context and Factuality. Artificial Analysis tested Sonnet 5.5 as Anthropic serves it by default, with its safeguard fallback model on. Artificial Analysis Computer use. Anthropic's own run at max effort in its computer-use harness, averaged over five runs. It fully completed 43.5% of the tasks. Anthropic Answers in full. SpeechMap tested it through OpenRouter at medium thinking and counted 13.5% of requests denied and 4.6% evaded. SpeechMap.AI
Claude Opus 5.5	65.2%	66.9%	91.2%	90.6%	61.4%	69.5%	81.8%Reported by Anthropic	84.7%	87.7%	65.5%	82.8%	Not yet published
Agentic coding. Vals.ai ranked it 1st on its leaderboard of Oct 24, 2026. Vals.ai Code, Hard reasoning, Tool use, Long context, Multimodal and Factuality. Artificial Analysis tested Opus 5.5 as Anthropic serves it by default, with its safeguard fallback model on. Artificial Analysis Computer use. Anthropic's own run at max effort in its computer-use harness, averaged over five runs. It fully completed 48.7% of the tasks. Anthropic Answers in full. SpeechMap tested it through OpenRouter at medium thinking and counted 12.7% of requests denied and 2.2% evaded. Another 2.3% failed at the provider and count against the score. SpeechMap.AI
Claude Fable 5.1	58.1%	63.1%	90.2%	93.4%Run by Vals.ai	59.1%	59.4%	80.7%Reported by Anthropic	85.3%	86.3%Run by Moonshot AI	68.4%	72.4%	Not yet published
Code, Hard reasoning, Tool use, Long context and Factuality. Artificial Analysis tested Fable 5.1 as Anthropic serves it by default, with its safeguard fallback model on. Artificial Analysis Science. Vals.ai's run at max effort, ranked 9th on its GPQA Diamond board, last updated September 1, 2026, before Vals archived the board. Vals.ai Computer use. Anthropic's own run on the v2.1 tasks, averaged over five runs. It replaces the 77.9% in the Fable 5.1 system card, which used the August 2026 task set. Anthropic Multimodal. Moonshot AI's run at max effort with Anthropic's fallback on, from the Kimi K3.1 Pro launch table, by the official MMMU-Pro protocol. Artificial Analysis has not run it. Moonshot AI Answers in full. With reasoning on, SpeechMap counted 15.1% of requests denied. Another 11.1% failed at the provider and count against the score, far more than any other Claude release. SpeechMap.AI
Claude Opus 5	53.5%	56.4%	85.6%	93.9%	54.9%	56.6%	77.7%	79.3%	84.7%	60.9%	73.7%	Not yet published
Computer use. The OSWorld team's own run at max effort, with its batch tool and a 500-step budget. It fully completed 44.33% of tasks. OSWorld team Answers in full. With reasoning on, SpeechMap counted 22.3% of requests denied and 3.5% evaded. With reasoning off, the model completes 59.1%. SpeechMap.AI
Claude Sonnet 5	10.6%	54.3%	65.6%	80.3%	41.3%	36.5%	57%Reported by Anthropic	82%	77.3%	40.1%	Not yet published	Not yet published
Computer use. Anthropic reran it for the Sonnet 5.5 launch, averaged over five runs. The share of tasks fully completed is 25.6%. Anthropic
Claude Fable 5	41.4%	61%	87%	85.9%	55.5%	54.1%	Not yet published	82.3%	Not yet published	65.4%	64.8%	63.5%Run by Artificial AnalysisIFBench
Code, Hard reasoning, Tool use, Long context and Factuality. Artificial Analysis tested Fable 5 with its Opus 4.8 fallback switched on, the setup its model page lists. Artificial Analysis Answers in full. With reasoning on, SpeechMap counted 27.9% of requests denied and 5.0% evaded. Another 2.4% failed at the provider and count against the score. SpeechMap.AI Instructions. Artificial Analysis ran Claude Fable 5 at max effort in its own evaluation harness. The run routes some requests to Opus 4.8 as a fallback. Artificial Analysis
Claude Opus 4.8	23.2%	54.4%	80%	91%	48.7%	45.6%	54.8%Run by OSWorld teamOSWorld 2.0	77.7%	Not yet published	48.8%	67.4%	62.2%Run by Artificial AnalysisIFBench
Computer use. The OSWorld team's own run at max effort with its batched tool and a 500-step budget, on the v2026.08.08 task release. It fully completed 20.6% of tasks; the standard tool setting scored 49.33. OSWorld team Answers in full. With thinking on, SpeechMap counted 25.2% of requests denied and 7.0% evaded. With thinking off, the model completes 43.4%. SpeechMap.AI Instructions. Artificial Analysis ran Claude Opus 4.8 at max effort in its own evaluation harness. Artificial Analysis
Claude Opus 4.7	Not yet published	54.5%	70.2%	86.4%	42.3%	Not yet published	48.9%Run by OSWorld teamOSWorld 2.0	78.7%	78.8%	48.9%	70.2%	58.6%Run by Artificial AnalysisIFBench
Computer use. The OSWorld team's own run at max effort with its batched tool and a 500-step budget, on the v2026.08.08 task release. It fully completed 18.2% of tasks; the standard tool setting scored 49.1. OSWorld team Answers in full. With reasoning on, SpeechMap counted 22.0% of requests denied and 5.1% evaded. With reasoning off, the model completes 71.5%. SpeechMap.AI Instructions. Artificial Analysis ran Claude Opus 4.7 at max effort in its own evaluation harness. Artificial Analysis
Claude Sonnet 4.6	3%Run by Artificial AnalysisTerminal-Bench 4.0	50.1%	71.1%Run by Epoch AIOTIS Mock AIME 2024-2025	78.8%	33.6%	20.1%	41.5%Run by OSWorld teamOSWorld 2.0	80%	73.3%	40.9%	35.8%	56.6%Run by Artificial AnalysisIFBench
Agentic coding. Artificial Analysis's run of the October 29 build at high effort in its own agent harness. Vals.ai has run only the October 6 preview build. Artificial Analysis Agentic coding. Unusual. Artificial Analysis's run of the October 29 build at high effort in its own agent harness. Vals.ai has run only the October 6 preview build. Oct 28, 2026 Artificial Analysis Math. Epoch AI's run at max effort, the highest effort it ran. Its 32K thinking budget run scored 85.8, and high 75.6. Epoch AI Computer use. The OSWorld team's own run at max effort with a 500-step budget, on the v2026.08.08 task release. It fully completed 8.3% of tasks; its medium-effort run scored 33.9. OSWorld team Answers in full. With thinking on, SpeechMap counted 45.4% of requests denied and 18.3% evaded. With thinking off, the model completes 29.1%. SpeechMap.AI Instructions. Artificial Analysis ran Claude Sonnet 4.6 at max effort in its own evaluation harness. Artificial Analysis
Claude Opus 4.6	Not yet published	Not yet published	66%	88.4%	39.9%	Not yet published	Not yet published	78%	75.4%	47%	59.9%	37.2%
Answers in full. With thinking on, SpeechMap counted 27.9% of requests denied and 10.4% evaded. With thinking off, the model completes 49.8%. SpeechMap.AI Instructions. On Oct 28, 2026, Scale AI ranked this run 35th. The same model with thinking off scores 56.02 on Scale's board. Scale AI SEAL
Claude Opus 4.5	Not yet published	Not yet published	34.4%	86%	30.1%	Not yet published	Not yet published	77.3%	74%	46.6%	57.5%	59%
Math. Epoch AI's run with a 32K-token thinking budget. Epoch AI Science. Epoch AI's run with a 32K-token thinking budget. With thinking off it scores 80.7%. Epoch AI Answers in full. With extended thinking on, SpeechMap counted 29.1% of requests denied and 13.3% evaded. With it off, the model completes 50.8%. SpeechMap.AI Instructions. On Oct 28, 2026, Scale AI ranked it 18th, with extended thinking on. Scale AI SEAL
Claude Haiku 4.5	Not yet published	42.2%	66.7%Run by Epoch AIOTIS Mock AIME 2024-2025	71.2%	10.4%	3.2%	Not yet published	74.3%	58.6%	18%	24.7%	50.5%
Math. Epoch AI's run with a 32K thinking budget, the highest it ran. Without extended thinking the same model scored 35.8. Epoch AI Science. Epoch AI's run with a 32K-token thinking budget. With thinking off it scores 60.5%. Epoch AI Tool use. Artificial Analysis publishes 3.2 for the reasoning version of Haiku 4.5 on this test, far below the model's other results. Artificial Analysis Tool use. Unusual. Artificial Analysis publishes 3.2 for the reasoning version of Haiku 4.5 on this test, far below the model's other results. Oct 28, 2026 Artificial Analysis Answers in full. With extended thinking on, SpeechMap counted 39.0% of requests denied and 36.4% evaded. With it off, the model completes 20.8%. SpeechMap.AI Instructions. On Oct 28, 2026, Scale AI ranked it 29th, with extended thinking on. Scale AI SEAL
Claude Sonnet 4.5	Not yet published	45.7%	23.9%	82.3%	17.8%	14.1%	Not yet published	72.3%	68.7%	32.9%	31.4%	55.3%
Math. Epoch AI's run with a 32K-token thinking budget. Epoch AI Science. Epoch AI's run with a 59K-token thinking budget. With thinking off it scores 73.7. Epoch AI Answers in full. With reasoning on, SpeechMap counted 48.3% of requests denied and 20.2% evaded. With reasoning off, the model completes 32.3%. SpeechMap.AI Instructions. On Oct 28, 2026, Scale AI ranked it 25th, with extended thinking on. Scale AI SEAL
Gemini 4 Argon	57.6%	61.8%	86%	96.4%	57.1%	77.5%	69.2%Reported by GoogleOSWorld 2.0	79.7%	Not yet published	48.3%	87%	78.7%
Computer use. Google's own run at its highest thinking setting in its Gemini computer-use harness, best of 3 runs. No outside lab has run it. Google Answers in full. SpeechMap tested it through OpenRouter at medium reasoning and counted 10.2% of requests denied and 2.4% evaded. SpeechMap.AI Instructions. On Oct 28, 2026, Scale AI ranked it 2nd. Scale AI SEAL
Gemini 3.8 Flash	19.7%	56.6%	68.4%	95.4%	47.8%	59.9%	59%Reported by GoogleOSWorld 2.0	81.3%	85.6%	50.5%	86.1%	Not yet published
Computer use. Google's own run in its Gemini computer-use harness, best of 3 runs, on the task files before the August patch. Google does not print the thinking level. Google Answers in full. SpeechMap tested it through OpenRouter with reasoning on and counted 11.7% of requests denied and 1.8% evaded. SpeechMap.AI
Gemma 4 31B	Not yet published	45.5%	Not yet published	85.7%Run by Artificial AnalysisGPQA Diamond	23.6%	4.9%	Not yet published	69.7%	73.4%	10.4%Run by Epoch AISimpleQA Verified	82.3%	75.6%Run by Artificial AnalysisIFBench
Science. Artificial Analysis's run of the reasoning version. Epoch AI, the axis's source, ran it only at minimal effort, which is not drawn. Artificial Analysis Tool use. Artificial Analysis publishes 4.9 for the reasoning version of Gemma 4 31B on this test, far below its other results. Artificial Analysis Tool use. Unusual. Artificial Analysis publishes 4.9 for the reasoning version of Gemma 4 31B on this test, far below its other results. Oct 28, 2026 Artificial Analysis Factuality. Epoch AI ran SimpleQA Verified on the instruction-tuned release. SimpleQA Verified prints no hallucination rate, so none is shown beside it. Epoch AI Answers in full. SpeechMap counted 16.8% of requests denied and 0.9% evaded. SpeechMap.AI Instructions. Artificial Analysis's run of the reasoning version on IFBench, a test of precise output constraints. Scale AI has not run it on MultiChallenge. Artificial Analysis
Gemini 3.1 Pro Preview	2.5%	58.7%	59.6%	94.4%	47%	35.4%	Not yet published	82%	82.4%	54.9%	59.9%	71.4%
Agentic coding. Vals.ai tested the February 2026 preview build. Vals.ai Agentic coding. Unusual. Vals.ai tested the February 2026 preview build. Oct 24, 2026 Vals.ai Long context. Artificial Analysis tested the February 19, 2026 preview build. Artificial Analysis Answers in full. With reasoning on, SpeechMap counted 30.9% of requests denied and 8.8% evaded. SpeechMap.AI Instructions. On Oct 28, 2026, Scale AI ranked it 8th. Scale does not say which reasoning setting it tested. Scale AI SEAL
Gemini 3.6 Flash	7.1%Run by Artificial AnalysisTerminal-Bench 4.0	53.4%	58.9%	94.1%	40.8%	53%	83%Reported by GoogleOSWorld-Verified	80%	83.2%	66.2%Run by Epoch AISimpleQA Verified	87.2%	Not yet published
Agentic coding. Artificial Analysis ran Terminal-Bench 4.0 in its own agent harness at high effort. Artificial Analysis Computer use. Google's own figure from the Gemini 3.6 Flash model card, results as of July 2026. It does not state the harness, step budget or thinking level. No independent lab has run it. Google Factuality. Epoch AI's run at high effort, the highest it ran. SimpleQA Verified prints no hallucination rate, so none is shown beside it. Epoch AI Answers in full. SpeechMap tested Gemini 3.6 Flash through its reasoning-on entry and reports complete, evasive, denial, and provider error shares. SpeechMap.AI
DeepSeek V4.1 Flash	31.8%	55.9%	71.2%	93.2%	43.3%	66.6%	69.3%Run by Moonshot AIOSWorld 2.1	82.7%	80.6%	39.6%	Not yet published	74.9%
Agentic coding. Vals.ai ranked the October 17 build 19th on its leaderboard of Oct 24, 2026. Vals.ai Computer use. Moonshot AI's run of the October 17 build at max effort in its Kimi Agent harness, from the Kimi K3.1 Pro launch table. Moonshot does not name the metric. Moonshot AI Instructions. On Oct 28, 2026, Scale AI ranked the October 17 build 5th. Scale AI SEAL
DeepSeek V4 Pro	14.1%	51%	64.6%	91.7%	41%	56.7%	Not yet published	80.3%	Not yet published	45.2%	72%	Not yet published
Agentic coding. Vals.ai ranked it 38th of 58 on its leaderboard of Oct 29, 2026. Vals.ai Factuality. Of the questions it did not know, it answered 94.1% wrongly when it should have declined. Artificial Analysis prints this hallucination rate beside accuracy. Artificial Analysis Answers in full. SpeechMap counted 27.6% of requests denied and 0.3% evaded. SpeechMap.AI
Kimi K3.1 Pro	63.6%	60.8%	Not yet published	94.2%Reported by Moonshot AI	52.5%	73.3%	79.6%Reported by Moonshot AIOSWorld 2.1	86.3%	83.9%	48.9%	Not yet published	Not yet published
Science. Moonshot AI's own figure, at its highest reasoning setting with no tools. No independent lab has run it yet. Epoch AI scores the earlier Kimi K3 at 93.1. Moonshot AI Computer use. Moonshot AI's own run at max effort in its Kimi Agent harness. Moonshot does not name the metric or task set; its other rows reprint Anthropic's full-set partial scores. Moonshot AI
Kimi K3	17.2%	59.5%	72.2%	93.1%	46.9%	58.3%	Not yet published	88.7%	80.5%	50.6%Run by Epoch AISimpleQA Verified	65.9%	Not yet published
Factuality. Epoch AI's run at max effort. SimpleQA Verified prints no hallucination rate, so none is shown beside it. Epoch AI Answers in full. SpeechMap tested Kimi K3 through its reasoning-on entry and reports complete, evasive, denial, and provider error shares. SpeechMap.AI
Kimi K2.7 Code	1%Run by Artificial AnalysisTerminal-Bench 4.0	47.8%	54%	87.9%	35%	24.5%	Not yet published	79.3%	Not yet published	36.5%Run by Epoch AISimpleQA Verified	Not yet published	63.1%Run by Artificial AnalysisIFBench
Agentic coding. Artificial Analysis ran Terminal-Bench 4.0 in its own agent harness at reasoning on effort. Artificial Analysis Agentic coding. Unusual. Artificial Analysis ran Terminal-Bench 4.0 in its own agent harness at reasoning on effort. Oct 31, 2026 Artificial Analysis Factuality. Epoch AI's run; Epoch prints no effort for it. SimpleQA Verified prints no hallucination rate, so none is shown beside it. Epoch AI Instructions. Artificial Analysis ran IFBench directly at reasoning on effort. Artificial Analysis
Mistral Large 4	24.7%Run by Artificial AnalysisTerminal-Bench 4.0	Not yet published	Not yet published	Not yet published	Not yet published	57.8%	Not yet published	Not yet published	Not yet published	26%	Not yet published	Not yet published
Agentic coding. Artificial Analysis's run of the October 29 build at high effort in its own agent harness. Vals.ai has run only the October 6 preview build. Artificial Analysis Tool use and Factuality. Artificial Analysis's run of the October 29 build at high effort, its highest run of that build. Artificial Analysis
Mistral Medium 3.5	0%Run by Artificial AnalysisTerminal-Bench 4.0	40.2%	Not yet published	74.8%Run by Artificial AnalysisGPQA Diamond	13.8%	6.3%	Not yet published	69.3%	64.9%	23%	70.5%	68.8%Run by Artificial AnalysisIFBench
Agentic coding. Artificial Analysis's run at high effort in its own agent harness, from its Intelligence Index evaluations. It solved none of the tasks. Vals.ai has not run it. Artificial Analysis Agentic coding. Unusual. Artificial Analysis's run at high effort in its own agent harness, from its Intelligence Index evaluations. It solved none of the tasks. Vals.ai has not run it. Oct 28, 2026 Artificial Analysis Science. Artificial Analysis's run at high effort, the highest it ran for this release, from its Intelligence Index evaluations. Epoch AI has not run it. Artificial Analysis Answers in full. SpeechMap tested it through OpenRouter with reasoning on. With reasoning off it answers 87.9% in full. SpeechMap.AI Instructions. Artificial Analysis's IFBench run at high effort: the share of out-of-domain output constraints the model keeps. Scale AI has not run MultiChallenge on it. Artificial Analysis
Grok 4.7	28.8%	57.4%	53%	92.7%	43.1%	65.6%	Not yet published	76.7%	Not yet published	42.7%	64.1%	Not yet published
Answers in full. SpeechMap tested it through OpenRouter at medium reasoning and counted 34.4% of requests denied and 1.4% evaded. SpeechMap.AI
Grok 4.6	17.2%	56.5%	66%	93.2%	42.9%	66.7%	Not yet published	80.3%	Not yet published	48.9%Run by Epoch AISimpleQA Verified	83.6%	Not yet published
Factuality. Epoch AI's run at xhigh effort, the highest it ran; its high run scored 49.3. SimpleQA Verified prints no hallucination rate. Epoch AI Answers in full. SpeechMap tested Grok 4.6 through its reasoning-on entry and reports complete, evasive, denial, and provider error shares. SpeechMap.AI
Grok 4.5	8.6%	55%	57.2%	93.4%	42.7%	57.9%	Not yet published	79.3%	80.4%	48.3%Run by Epoch AISimpleQA Verified	93.1%	Not yet published
Factuality. Epoch AI's run at high effort, the highest it ran. SimpleQA Verified prints no hallucination rate, so none is shown beside it. Epoch AI Answers in full. SpeechMap tested Grok 4.5 through its reasoning-on entry and reports complete, evasive, denial, and provider error shares. SpeechMap.AI
MiniMax M3.1 Flash	25.3%	53.4%	Not yet published	Not yet published	48%	72.5%	61.7%Reported by MiniMaxOSWorld 2.0	79.7%	79.1%	45.1%	Not yet published	Not yet published
Computer use. MiniMax's own figure on the August 2026 full task release, from its model card. MiniMax names neither the metric nor the harness nor the reasoning setting. MiniMax
Mercury 3	3%	Not yet published	Not yet published	Not yet published	Not yet published	Not yet published	Not yet published	Not yet published	Not yet published	Not yet published	Not yet published	Not yet published
GLM-5.3-Flash	37.4%	51.6%	55.8%	90.2%	39.9%	60.4%	59.1%Reported by Z.aiOSWorld 2.0	80%	86%Run by Vals.aiMMMU-Pro	28.2%	65.4%	Not yet published
Computer use. Z.ai's own figure from its launch post. It does not name the OSWorld 2.0 task release, task set, harness or metric. No independent lab has run it. Z.ai Multimodal. Vals.ai's run, rank 23 on its MMMU Pro board, last updated September 1, 2026. Vals's variant offers 4 answer options, so it reads higher than the 10-option set. Vals.ai Answers in full. SpeechMap counted 28.9% of requests denied and 3.3% evaded. 2.4% failed at the provider. SpeechMap.AI
GLM-5.3	38.9%	59%	68.8%	90.9%	42.3%	62.2%	Not yet published	79.7%	Not yet published	32.3%	71.5%	Not yet published
Answers in full. SpeechMap counted 25.1% of requests denied and 3.3% evaded. SpeechMap.AI
GLM-5.2	1%Run by Artificial AnalysisTerminal-Bench 4.0	51.2%	59.2%	Not yet published	41.1%	28.4%	Not yet published	78.3%	Not yet published	34.2%Run by Epoch AISimpleQA Verified	Not yet published	73.3%Run by Artificial AnalysisIFBench
Agentic coding. Artificial Analysis ran Terminal-Bench 4.0 in its own agent harness at max effort. Artificial Analysis Agentic coding. Unusual. Artificial Analysis ran Terminal-Bench 4.0 in its own agent harness at max effort. Oct 31, 2026 Artificial Analysis Factuality. Epoch AI's run at max effort. SimpleQA Verified prints no hallucination rate, so none is shown beside it. Epoch AI Instructions. Artificial Analysis ran IFBench directly at max effort. Artificial Analysis
Muse Spark Open	6.6%	58.4%	69.8%	88.6%	43.9%	49.7%	80.4%Reported by MetaOSWorld-Verified	78.3%	80.9%	42%	Not yet published	74.2%
Agentic coding. Vals.ai ranked it 47th of 58 on its leaderboard of Oct 29, 2026. Artificial Analysis's own-harness run agrees, at 8.1. Vals.ai Agentic coding. Unusual. Vals.ai ranked it 47th of 58 on its leaderboard of Oct 29, 2026. Artificial Analysis's own-harness run agrees, at 8.1. Oct 29, 2026 Vals.ai Computer use. Meta's own figure from the Muse Spark Open model card, at xhigh effort. Meta does not state the harness or step budget. Meta Instructions. On Oct 29, 2026, Scale AI ranked it 6th, level with GPT-6.1 Terra. Scale AI SEAL
Muse Wonder	49.5%	60.3%	82.1%	Not yet published	50.1%	68.1%	72.6%Reported by Meta	80.7%	Not yet published	38%	42.7%	Not yet published
Agentic coding. Vals.ai ranked it 12th on its leaderboard of Oct 29, 2026, level with GPT-6.1 Terra at 49.49%. Vals.ai Computer use. Meta's own run at max effort in its own test framework on all 108 tasks: the average partial score. It fully completed 34.3% of tasks. Meta Answers in full. SpeechMap tested it through OpenRouter with reasoning on. The provider's API blocked 54.8% of requests; the model denied 1.4% and evaded 1.1%. SpeechMap.AI
Muse Spark 1.3	10.6%	58.8%	74%	93.5%Run by Artificial AnalysisGPQA Diamond	48.7%	57.9%	69.6%	83%	82%	38%	25.8%	Not yet published
Science. Artificial Analysis's run at max effort, the highest it ran for this release, from its Intelligence Index evaluations; its xhigh run scored 94.1. Epoch AI has not run it. Artificial Analysis Computer use. The OSWorld team's own run at max effort, with its batch tool and a 500-step budget. It fully completed 39.25% of tasks. OSWorld team Multimodal. Artificial Analysis ran MMMU-Pro on this release at xhigh effort only; its Max record carries no MMMU-Pro figure. Artificial Analysis Answers in full. SpeechMap tested Muse Spark 1.3 through its reasoning-on entry and reports complete, evasive, denial, and provider error shares. SpeechMap.AI Answers in full. Unusual. SpeechMap tested Muse Spark 1.3 through its reasoning-on entry and reports complete, evasive, denial, and provider error shares. Oct 31, 2026 SpeechMap.AI
Llama 4 Scout	Not yet published	21.3%	7.8%Run by Epoch AIOTIS Mock AIME 2024-2025	51.8%	3.8%	0.2%	Not yet published	27.7%	52.9%	Not yet published	70.1%	39.5%Run by Artificial AnalysisIFBench
Math. Epoch AI's run of the instruction-tuned release. Llama 4 has no reasoning mode. Epoch AI Hard reasoning. Artificial Analysis publishes 3.8 for Llama 4 Scout, which has no reasoning mode, on this test. Artificial Analysis Tool use. Artificial Analysis publishes 0.2 for Llama 4 Scout on this test, far below its other results. Artificial Analysis Instructions. Artificial Analysis ran Llama 4 Scout, which has no reasoning mode, in its own evaluation harness. Artificial Analysis
Llama 4 Maverick	Not yet published	31.7%	20.6%Run by Epoch AIOTIS Mock AIME 2024-2025	67%	4.9%	0.2%	Not yet published	50%	62.1%	Not yet published	58.3%	43%Run by Artificial AnalysisIFBench
Math. Epoch AI's run of the FP8 instruction-tuned release. Llama 4 has no reasoning mode. Epoch AI Science. Epoch AI ran Meta's FP8 release of the weights. Epoch AI Hard reasoning. Artificial Analysis publishes 4.9 for Llama 4 Maverick, which has no reasoning mode, on this test. Artificial Analysis Hard reasoning. Unusual. Artificial Analysis publishes 4.9 for Llama 4 Maverick, which has no reasoning mode, on this test. Oct 28, 2026 Artificial Analysis Tool use. Artificial Analysis publishes 0.2 for Llama 4 Maverick on this test, far below its other results. Artificial Analysis Tool use. Unusual. Artificial Analysis publishes 0.2 for Llama 4 Maverick on this test, far below its other results. Oct 28, 2026 Artificial Analysis Instructions. Artificial Analysis ran Llama 4 Maverick, which has no reasoning mode, in its own evaluation harness. Artificial Analysis
Qwen3.8 27B	5.6%Run by Artificial AnalysisTerminal-Bench 4.0	46.6%	Not yet published	90.5%Run by Artificial AnalysisGPQA Diamond	33.9%	48.2%	84.3%Reported by AlibabaOSWorld-Verified	82%	76.3%	17.3%	Not yet published	79.5%Reported by AlibabaIFBench
Agentic coding. Artificial Analysis's run at xhigh effort in its own agent harness, from its Intelligence Index evaluations. Vals.ai has not run it. Artificial Analysis Agentic coding. Unusual. Artificial Analysis's run at xhigh effort in its own agent harness, from its Intelligence Index evaluations. Vals.ai has not run it. Nov 2, 2026 Artificial Analysis Science. Artificial Analysis's run at xhigh effort, the highest it ran for this release, from its Intelligence Index evaluations. Epoch AI has not run it. Artificial Analysis Computer use. Alibaba's own figure from the Qwen3.8 27B model card. It does not state the harness, step budget or reasoning effort. Alibaba Instructions. Alibaba's own figure from the Qwen3.8 27B model card. It does not state the reasoning effort. Scale AI has not run MultiChallenge on it. Alibaba
Qwen3.8 Max	38.9%	52.1%	65.6%	92.3%	43.1%	56.2%	Not yet published	80.3%	82.8%	29.5%	65.3%	Not yet published
Agentic coding. Vals.ai ranked the September 2 build 16th on its leaderboard of Oct 29, 2026. Vals.ai Answers in full. SpeechMap tested the September 2 build with reasoning on and counted 29.8% of requests denied and 2.1% evaded. SpeechMap.AI
MiMo-V2.6-Pro	31.3%	60.9%	Not yet published	Not yet published	49.4%	58.6%	82%Reported by XiaomiOSWorld-Verified	86.3%	Not yet published	34.8%	66.5%	Not yet published
Computer use. Xiaomi's own figure from the MiMo-V2.6-Pro model card. It does not state the harness, step budget or reasoning effort. No independent lab has run it. Xiaomi Answers in full. SpeechMap counted 29.7% of requests denied and 3.8% evaded. SpeechMap.AI
MiMo-V2.6-Flash	24.2%	51.3%	Not yet published	Not yet published	35.1%	64.1%	80.8%Reported by XiaomiOSWorld-Verified	74.3%	73.1%	27.2%	66.9%	Not yet published
Computer use. Xiaomi's own figure from the MiMo-V2.6-Flash model card. It does not state the harness, step budget or reasoning effort. No independent lab has run it. Xiaomi Answers in full. SpeechMap counted 30.1% of requests denied and 3.0% evaded. SpeechMap.AI
Ling 3.1 Flash	32.8%Run by Artificial AnalysisTerminal-Bench 4.0	56.8%	Not yet published	Not yet published	43.5%	67.5%	Not yet published	82.3%	Not yet published	26.5%	Not yet published	Not yet published
Agentic coding. Artificial Analysis's run with reasoning on in its own agent harness, from its Intelligence Index evaluations. Vals.ai has not run it. Artificial Analysis
Step 5 Preview	31.8%	58.9%	Not yet published	93.5%Reported by StepFunGPQA Diamond	46.5%	51%	Not yet published	88.3%	76.4%	36.4%	59.7%	Not yet published
Agentic coding. Vals.ai ranked it 22nd of 58 on its leaderboard of Oct 29, 2026. Vals.ai Science. StepFun's own figure from its launch table, at High reasoning, its highest setting. Epoch AI has not run it. StepFun Answers in full. SpeechMap ran it at medium reasoning through OpenRouter: 34.5% denied, 1.0% evaded, and 4.8% stopped by the provider's content filter. SpeechMap.AI

Axis

Agentic codingTerminal-Bench 4.0

Claude Opus 5.565.2%
Claude Sonnet 5.564.1%
Kimi K3.1 Pro63.6%
GPT-6 Astra59.6%
Claude Fable 5.158.1%
Gemini 4 Argon57.6%
GPT-6.1 Sol55.1%
Claude Opus 553.5%
GPT-6.1 Terra49.5%
Muse Wonder49.5%
GPT-6 Sol44.4%
Claude Fable 541.4%
GLM-5.338.9%
Qwen3.8 Max38.9%
GPT-5.6 Sol37.9%
GLM-5.3-Flash37.4%
Claude Haiku 5.536.4%
DeepSeek V4.1 Flash31.8%
Step 5 Preview31.8%
MiMo-V2.6-Pro31.3%
Grok 4.728.8%
MiniMax M3.1 Flash25.3%
MiMo-V2.6-Flash24.2%
Claude Opus 4.823.2%
GPT-5.6 Terra22.7%
Gemini 3.8 Flash19.7%
Kimi K317.2%
Grok 4.617.2%
DeepSeek V4 Pro14.1%
GPT-6 Luna13.6%
GPT-5.6 Luna11.6%
Claude Sonnet 510.6%
Muse Spark 1.310.6%
Grok 4.58.6%
Muse Spark Open6.6%
Mercury 33%
Gemini 3.1 Pro Preview2.5%
Ling 3.1 Flash32.8%Artificial Analysis's run, Terminal-Bench 4.0
Mistral Large 424.7%Artificial Analysis's run, Terminal-Bench 4.0
GPT-5.514.6%Artificial Analysis's run, Terminal-Bench 4.0
gpt-oss-120b14.2%Terminal-Bench's run, Terminal-Bench 2.0
Gemini 3.6 Flash7.1%Artificial Analysis's run, Terminal-Bench 4.0
Qwen3.8 27B5.6%Artificial Analysis's run, Terminal-Bench 4.0
Claude Sonnet 4.63%Artificial Analysis's run, Terminal-Bench 4.0
Kimi K2.7 Code1%Artificial Analysis's run, Terminal-Bench 4.0
GLM-5.21%Artificial Analysis's run, Terminal-Bench 4.0
Mistral Medium 3.50%Artificial Analysis's run, Terminal-Bench 4.0

The other models in the grid have no result on Terminal-Bench 4.0 yet.

CodeSciCode

The other models in the grid have no result on SciCode yet.

MathFrontierMath Tiers 1-3

GPT-6.1 Sol93.7%
GPT-6 Astra93.7%
Claude Opus 5.591.2%
GPT-6.1 Terra90.9%
Claude Fable 5.190.2%
GPT-6 Sol89.8%
GPT-5.6 Sol89.1%
Claude Sonnet 5.588.8%
Claude Fable 587%
GPT-5.6 Terra86%
Gemini 4 Argon86%
Claude Opus 585.6%
GPT-5.585.3%
GPT-5.6 Luna82.1%
Muse Wonder82.1%
Claude Opus 4.880%
GPT-6 Luna78.9%
GPT-5.478.6%
Claude Haiku 5.575.1%
Muse Spark 1.374%
Kimi K372.2%
DeepSeek V4.1 Flash71.2%
Claude Opus 4.770.2%
Muse Spark Open69.8%
GLM-5.368.8%
Gemini 3.8 Flash68.4%
Claude Opus 4.666%
Grok 4.666%
Claude Sonnet 565.6%
Qwen3.8 Max65.6%
DeepSeek V4 Pro64.6%
Gemini 3.1 Pro Preview59.6%
GLM-5.259.2%
Gemini 3.6 Flash58.9%
Grok 4.557.2%
GLM-5.3-Flash55.8%
Kimi K2.7 Code54%
Grok 4.753%
Claude Opus 4.534.4%
Claude Sonnet 4.523.9%
gpt-oss-120b88.9%Epoch AI's run, OTIS Mock AIME 2024-2025
Claude Sonnet 4.671.1%Epoch AI's run, OTIS Mock AIME 2024-2025
Claude Haiku 4.566.7%Epoch AI's run, OTIS Mock AIME 2024-2025
Llama 4 Maverick20.6%Epoch AI's run, OTIS Mock AIME 2024-2025
Llama 4 Scout7.8%Epoch AI's run, OTIS Mock AIME 2024-2025

The other models in the grid have no result on FrontierMath Tiers 1-3 yet.

ScienceGPQA Diamond

Gemini 4 Argon96.4%
GPT-6 Astra95.8%
Claude Sonnet 5.595.6%
GPT-6.1 Sol95.4%
Gemini 3.8 Flash95.4%
Gemini 3.1 Pro Preview94.4%
GPT-6.1 Terra94.3%
GPT-6 Sol94.3%
Gemini 3.6 Flash94.1%
Claude Opus 593.9%
GPT-5.6 Sol93.5%
Grok 4.593.4%
GPT-5.6 Terra93.3%
GPT-5.493.3%
DeepSeek V4.1 Flash93.2%
Grok 4.693.2%
Kimi K393.1%
Grok 4.792.7%
Qwen3.8 Max92.3%
DeepSeek V4 Pro91.7%
GPT-5.6 Luna91.6%
Claude Opus 4.891%
GLM-5.390.9%
Claude Opus 5.590.6%
GPT-6 Luna90.5%
GLM-5.3-Flash90.2%
Claude Haiku 5.589.6%
Muse Spark Open88.6%
Claude Opus 4.688.4%
Kimi K2.7 Code87.9%
Claude Opus 4.786.4%
Claude Opus 4.586%
Claude Fable 585.9%
Claude Sonnet 4.582.3%
Claude Sonnet 580.3%
Claude Sonnet 4.678.8%
gpt-oss-120b75.8%
Claude Haiku 4.571.2%
Llama 4 Maverick67%
Llama 4 Scout51.8%
Kimi K3.1 Pro94.2%Moonshot AI's own figure
GPT-5.593.5%Artificial Analysis's run
Muse Spark 1.393.5%Artificial Analysis's run, GPQA Diamond
Step 5 Preview93.5%StepFun's own figure, GPQA Diamond
Claude Fable 5.193.4%Vals.ai's run
GPT-5.3-Codex91.5%Artificial Analysis's run
Qwen3.8 27B90.5%Artificial Analysis's run, GPQA Diamond
Gemma 4 31B85.7%Artificial Analysis's run, GPQA Diamond
Mistral Medium 3.574.8%Artificial Analysis's run, GPQA Diamond

The other models in the grid have no result on GPQA Diamond yet.

Hard reasoningHumanity's Last Exam

The other models in the grid have no result on Humanity's Last Exam yet.

Tool useAutomationBench-AA

The other models in the grid have no result on AutomationBench-AA yet.

Computer useOSWorld 2.1

Claude Opus 577.7%
Muse Spark 1.369.6%
Qwen3.8 27B84.3%Alibaba's own figure, OSWorld-Verified
Gemini 3.6 Flash83%Google's own figure, OSWorld-Verified
MiMo-V2.6-Pro82%Xiaomi's own figure, OSWorld-Verified
Claude Opus 5.581.8%Anthropic's own figure
MiMo-V2.6-Flash80.8%Xiaomi's own figure, OSWorld-Verified
Claude Fable 5.180.7%Anthropic's own figure
Muse Spark Open80.4%Meta's own figure, OSWorld-Verified
Claude Sonnet 5.580.1%Anthropic's own figure
Kimi K3.1 Pro79.6%Moonshot AI's own figure, OSWorld 2.1
GPT-6.1 Sol78%OpenAI's own figure, OSWorld 2.1
GPT-6 Astra75.8%Meta's run
GPT-6.1 Terra74.2%OpenAI's own figure, OSWorld 2.1
Muse Wonder72.6%Meta's own figure
Claude Haiku 5.572.4%Anthropic's own figure, OSWorld 2.1
DeepSeek V4.1 Flash69.3%Moonshot AI's run, OSWorld 2.1
Gemini 4 Argon69.2%Google's own figure, OSWorld 2.0
GPT-6 Sol64.4%OpenAI's own figure, OSWorld 2.0
GPT-5.6 Sol62.7%OSWorld team's run, OSWorld 2.0
MiniMax M3.1 Flash61.7%MiniMax's own figure, OSWorld 2.0
GLM-5.3-Flash59.1%Z.ai's own figure, OSWorld 2.0
Gemini 3.8 Flash59%Google's own figure, OSWorld 2.0
Claude Sonnet 557%Anthropic's own figure
Claude Opus 4.854.8%OSWorld team's run, OSWorld 2.0
GPT-5.549.5%OSWorld team's run, OSWorld 2.0
GPT-6 Luna48.9%Anthropic's run, OSWorld 2.1
Claude Opus 4.748.9%OSWorld team's run, OSWorld 2.0
Claude Sonnet 4.641.5%OSWorld team's run, OSWorld 2.0

The other models in the grid have no result on OSWorld 2.1 yet.

Long contextAA-LCR v1.1

The other models in the grid have no result on AA-LCR v1.1 yet.

MultimodalMMMU-Pro

The other models in the grid have no result on MMMU-Pro yet.

FactualityAA-Omniscience Accuracy

Claude Fable 5.168.4%
Claude Opus 5.565.5%
Claude Fable 565.4%
GPT-6.1 Sol61.8%
GPT-6 Astra61.6%
Claude Opus 560.9%
Gemini 3.1 Pro Preview54.9%
GPT-6 Sol54.5%
Claude Sonnet 5.551.5%
Gemini 3.8 Flash50.5%
Claude Opus 4.748.9%
Kimi K3.1 Pro48.9%
Claude Opus 4.848.8%
Gemini 4 Argon48.3%
GPT-6.1 Terra47.6%
Claude Opus 4.647%
Claude Opus 4.546.6%
DeepSeek V4 Pro45.2%
MiniMax M3.1 Flash45.1%
Grok 4.742.7%
GPT-5.6 Terra42.3%
Muse Spark Open42%
Claude Sonnet 4.640.9%
Claude Sonnet 540.1%
GPT-6 Luna39.6%
DeepSeek V4.1 Flash39.6%
Muse Wonder38%
Muse Spark 1.338%
Step 5 Preview36.4%
MiMo-V2.6-Pro34.8%
Claude Haiku 5.534.7%
Claude Sonnet 4.532.9%
GLM-5.332.3%
Qwen3.8 Max29.5%
GLM-5.3-Flash28.2%
MiMo-V2.6-Flash27.2%
Ling 3.1 Flash26.5%
Mistral Large 426%
Mistral Medium 3.523%
Claude Haiku 4.518%
Qwen3.8 27B17.3%
GPT-5.6 Sol69.7%Epoch AI's run, SimpleQA Verified
Gemini 3.6 Flash66.2%Epoch AI's run, SimpleQA Verified
GPT-5.563%Epoch AI's run, SimpleQA Verified
Kimi K350.6%Epoch AI's run, SimpleQA Verified
Grok 4.648.9%Epoch AI's run, SimpleQA Verified
Grok 4.548.3%Epoch AI's run, SimpleQA Verified
GPT-5.6 Luna41%Epoch AI's run, SimpleQA Verified
Kimi K2.7 Code36.5%Epoch AI's run, SimpleQA Verified
GLM-5.234.2%Epoch AI's run, SimpleQA Verified
gpt-oss-120b16.8%OpenAI's own figure, SimpleQA
Gemma 4 31B10.4%Epoch AI's run, SimpleQA Verified

The other models in the grid have no result on AA-Omniscience Accuracy yet.

Instruction followingMultiChallenge

GPT-6.1 Sol82%
Gemini 4 Argon78.7%
DeepSeek V4.1 Flash74.9%
GPT-6.1 Terra74.2%
Muse Spark Open74.2%
Gemini 3.1 Pro Preview71.4%
Claude Haiku 5.567.7%
Claude Opus 4.559%
Claude Sonnet 4.555.3%
Claude Haiku 4.550.5%
gpt-oss-120b45.3%
Claude Opus 4.637.2%
Qwen3.8 27B79.5%Alibaba's own figure, IFBench
GPT-5.575.9%Artificial Analysis's run, IFBench
Gemma 4 31B75.6%Artificial Analysis's run, IFBench
GPT-5.3-Codex75.4%Artificial Analysis's run, IFBench
GPT-5.473.9%Artificial Analysis's run, IFBench
GLM-5.273.3%Artificial Analysis's run, IFBench
GPT-5.6 Sol72.7%Artificial Analysis's run, IFBench
GPT-5.6 Terra71.2%Artificial Analysis's run, IFBench
Mistral Medium 3.568.8%Artificial Analysis's run, IFBench
Claude Fable 563.5%Artificial Analysis's run, IFBench
Kimi K2.7 Code63.1%Artificial Analysis's run, IFBench
Claude Opus 4.862.2%Artificial Analysis's run, IFBench
Claude Opus 4.758.6%Artificial Analysis's run, IFBench
Claude Sonnet 4.656.6%Artificial Analysis's run, IFBench
Llama 4 Maverick43%Artificial Analysis's run, IFBench
Llama 4 Scout39.5%Artificial Analysis's run, IFBench

The other models in the grid have no result on MultiChallenge yet.

Releases

Every gpt-oss release we track, newest first.

gpt-oss-120bOpenAI's largest open-weight model, 117B parameters with 5.1B active, with a 131,072-token context window. Weights under the Apache 2.0 license; OpenAI lists no API price for it, so it runs on your own hardware or a host's.CurrentAug 5, 2025

What practitioners report

Strongest on Speed, weakest on Developer experience.

Each figure is the share of accounts naming gpt-oss-120b that speak well of it on that axis. An axis is scored once praise and concerns are both on record from at least five accounts by three authors. An account from the maker, a company that sells or serves the model, or the maker's expert program is listed and counts toward no figure.

Share of practitioner accounts favorable to gpt-oss-120b on each of seven axes, with the accounts behind each figure
Axis	Favorable	Accounts
Quality and reasoningWhether its answers and code are right on the author's own task: reasoning, accuracy, instruction following, and consistency.	58%	22 favorable, 16 not, 38 authors
Tool use and structured outputWhether it calls tools correctly, returns valid structured output, and keeps agent loops on track.	27%	4 favorable, 11 not, 15 authors
Long contextHow it holds up over long documents, large codebases, long sessions, and many images or PDFs.	25%	2 favorable, 6 not, 8 authors
Developer experienceHow it is to build with: prompting, steering, effort settings, and support across SDKs and providers.	14%	1 favorable, 6 not, 7 authors
CostWhat the work costs: price per task, token use, and what caching and batch save.	60%	3 favorable, 2 not, 5 authors
SpeedResponse time, time to first token, and throughput.	85%	23 favorable, 4 not, 27 authors
ReliabilityWhether requests keep succeeding: overloaded errors, outages, rate limits, and changes that broke what worked.	Only concerns on record	0 favorable, 4 not, 4 authors

Every model on these seven axes

Share of accounts favorable on each axis for gpt-oss-120b and every other model GILD has accounts for
Axis	gpt-oss-120b	GPT-6.1 Terra	GPT-6.1 Sol	GPT-6 Luna	GPT-6 Astra	GPT-5.6 Sol	GPT-5.6 Terra	GPT-5.6 Luna	GPT-5.5	GPT-5.4	GPT-6 Sol	GPT-5.3-Codex	Claude Haiku 5.5	Claude Sonnet 5.5	Claude Opus 5.5	Claude Fable 5.1	Claude Opus 5	Claude Sonnet 5	Claude Fable 5	Claude Opus 4.8	Claude Opus 4.7	Claude Sonnet 4.6	Claude Opus 4.6	Claude Opus 4.5	Claude Haiku 4.5	Claude Sonnet 4.5	Gemini 4 Argon	Gemini 3.8 Flash	Gemma 4 31B	Gemini 3.1 Pro Preview	Gemini 3.6 Flash	DeepSeek V4.1 Flash	DeepSeek V4 Pro	Kimi K3.1 Pro	Kimi K3	Kimi K2.7 Code	Mistral Large 4	Mistral Medium 3.5	Grok 4.7	Grok 4.6	Grok 4.5	MiniMax M3.1 Flash	Mercury 3	GLM-5.3-Flash	GLM-5.3	GLM-5.2	Muse Spark Open	Muse Wonder	Muse Spark 1.3	Llama 4 Scout	Llama 4 Maverick	Qwen3.8 27B	Qwen3.8 Max	MiMo-V2.6-Pro	MiMo-V2.6-Flash	Ling 3.1 Flash	Step 5 Preview
Quality and reasoning	58%22 favorable, 16 not, 38 authors	33%3 favorable, 6 not, 8 authors	55%12 favorable, 10 not, 22 authors	55%16 favorable, 13 not, 29 authors	44%12 favorable, 15 not, 26 authors	33%3 favorable, 6 not, 9 authors	38%3 favorable, 5 not, 8 authors	Needs 5 accounts from 3 authors3 favorable, 1 not, 4 authors	33%4 favorable, 8 not, 12 authors	33%2 favorable, 4 not, 6 authors	31%10 favorable, 22 not, 32 authors	Needs 5 accounts from 3 authors1 favorable, 3 not, 4 authors	73%19 favorable, 7 not, 26 authors	64%14 favorable, 8 not, 22 authors	68%17 favorable, 8 not, 25 authors	54%15 favorable, 13 not, 28 authors	50%6 favorable, 6 not, 12 authors	56%9 favorable, 7 not, 16 authors	68%13 favorable, 6 not, 18 authors	68%17 favorable, 8 not, 25 authors	57%12 favorable, 9 not, 21 authors	67%12 favorable, 6 not, 18 authors	70%19 favorable, 8 not, 27 authors	62%13 favorable, 8 not, 20 authors	58%14 favorable, 10 not, 24 authors	69%11 favorable, 5 not, 16 authors	63%22 favorable, 13 not, 35 authors	56%27 favorable, 21 not, 45 authors	80%24 favorable, 6 not, 28 authors	56%20 favorable, 16 not, 36 authors	Needs 5 accounts from 3 authors1 favorable, 3 not, 4 authors	61%25 favorable, 16 not, 40 authors	Only concerns on record0 favorable, 4 not, 4 authors	59%20 favorable, 14 not, 32 authors	Only praise on record3 favorable, 0 not, 3 authors	50%3 favorable, 3 not, 6 authors	58%11 favorable, 8 not, 18 authors	36%5 favorable, 9 not, 14 authors	43%16 favorable, 21 not, 36 authors	86%6 favorable, 1 not, 7 authors	60%3 favorable, 2 not, 5 authors	62%13 favorable, 8 not, 21 authors	86%18 favorable, 3 not, 21 authors	67%14 favorable, 7 not, 21 authors	76%47 favorable, 15 not, 61 authors	71%5 favorable, 2 not, 7 authors	Needs 5 accounts from 3 authors2 favorable, 1 not, 3 authors	20%1 favorable, 4 not, 5 authors	Only praise on record2 favorable, 0 not, 2 authors	53%8 favorable, 7 not, 15 authors	44%8 favorable, 10 not, 17 authors	71%24 favorable, 10 not, 34 authors	57%4 favorable, 3 not, 7 authors	22%2 favorable, 7 not, 9 authors	40%4 favorable, 6 not, 10 authors	Only concerns on record0 favorable, 1 not, 1 author	Needs 5 accounts from 3 authors2 favorable, 1 not, 3 authors
Tool use and structured output	27%4 favorable, 11 not, 15 authors	40%2 favorable, 3 not, 5 authors	40%4 favorable, 6 not, 10 authors	33%3 favorable, 6 not, 9 authors	62%8 favorable, 5 not, 13 authors	No accounts yet	Only concerns on record0 favorable, 2 not, 2 authors	Only praise on record1 favorable, 0 not, 1 author	Only praise on record1 favorable, 0 not, 1 author	No accounts yet	56%5 favorable, 4 not, 9 authors	No accounts yet	67%6 favorable, 3 not, 9 authors	40%4 favorable, 6 not, 10 authors	46%6 favorable, 7 not, 13 authors	30%3 favorable, 7 not, 10 authors	Needs 5 accounts from 3 authors3 favorable, 1 not, 4 authors	70%7 favorable, 3 not, 10 authors	57%4 favorable, 3 not, 7 authors	50%4 favorable, 4 not, 8 authors	50%4 favorable, 4 not, 8 authors	29%2 favorable, 5 not, 7 authors	33%4 favorable, 8 not, 12 authors	70%7 favorable, 3 not, 10 authors	43%3 favorable, 4 not, 6 authors	50%3 favorable, 3 not, 6 authors	25%2 favorable, 6 not, 8 authors	47%7 favorable, 8 not, 15 authors	67%4 favorable, 2 not, 6 authors	54%7 favorable, 6 not, 12 authors	Only concerns on record0 favorable, 1 not, 1 author	59%10 favorable, 7 not, 17 authors	Only concerns on record0 favorable, 1 not, 1 author	Needs 5 accounts from 3 authors3 favorable, 1 not, 4 authors	Only concerns on record0 favorable, 1 not, 1 author	No accounts yet	38%5 favorable, 8 not, 13 authors	Only concerns on record0 favorable, 5 not, 5 authors	38%6 favorable, 10 not, 16 authors	Only praise on record2 favorable, 0 not, 2 authors	Only concerns on record0 favorable, 1 not, 1 author	67%4 favorable, 2 not, 6 authors	80%8 favorable, 2 not, 10 authors	33%2 favorable, 4 not, 6 authors	57%4 favorable, 3 not, 7 authors	No accounts yet	No accounts yet	Only concerns on record0 favorable, 1 not, 1 author	Needs 5 accounts from 3 authors1 favorable, 1 not, 2 authors	Needs 5 accounts from 3 authors1 favorable, 2 not, 3 authors	Only concerns on record0 favorable, 4 not, 4 authors	46%6 favorable, 7 not, 13 authors	Only praise on record1 favorable, 0 not, 1 author	Only concerns on record0 favorable, 5 not, 5 authors	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 1 not, 1 author
Long context	25%2 favorable, 6 not, 8 authors	Only concerns on record0 favorable, 1 not, 1 author	38%3 favorable, 5 not, 8 authors	20%1 favorable, 4 not, 5 authors	27%3 favorable, 8 not, 11 authors	No accounts yet	No accounts yet	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 1 not, 1 author	Needs 5 accounts from 3 authors2 favorable, 1 not, 3 authors	Only concerns on record0 favorable, 3 not, 3 authors	No accounts yet	29%2 favorable, 5 not, 7 authors	60%3 favorable, 2 not, 5 authors	80%8 favorable, 2 not, 10 authors	63%5 favorable, 3 not, 8 authors	67%4 favorable, 2 not, 6 authors	50%3 favorable, 3 not, 6 authors	33%2 favorable, 4 not, 6 authors	60%6 favorable, 4 not, 10 authors	50%4 favorable, 4 not, 7 authors	71%5 favorable, 2 not, 7 authors	64%7 favorable, 4 not, 11 authors	40%4 favorable, 6 not, 10 authors	40%2 favorable, 3 not, 5 authors	50%4 favorable, 4 not, 8 authors	25%2 favorable, 6 not, 8 authors	38%5 favorable, 8 not, 13 authors	20%1 favorable, 4 not, 5 authors	44%7 favorable, 9 not, 16 authors	Only praise on record1 favorable, 0 not, 1 author	58%7 favorable, 5 not, 12 authors	No accounts yet	Needs 5 accounts from 3 authors3 favorable, 1 not, 4 authors	Only praise on record1 favorable, 0 not, 1 author	Only concerns on record0 favorable, 1 not, 1 author	Needs 5 accounts from 3 authors1 favorable, 3 not, 4 authors	No accounts yet	42%5 favorable, 7 not, 12 authors	No accounts yet	No accounts yet	40%2 favorable, 3 not, 5 authors	60%3 favorable, 2 not, 5 authors	No accounts yet	40%2 favorable, 3 not, 5 authors	No accounts yet	No accounts yet	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 3 not, 3 authors	Needs 5 accounts from 3 authors2 favorable, 1 not, 3 authors	75%6 favorable, 2 not, 8 authors	Only praise on record1 favorable, 0 not, 1 author	Needs 5 accounts from 3 authors1 favorable, 1 not, 2 authors	Needs 5 accounts from 3 authors1 favorable, 1 not, 2 authors	No accounts yet	No accounts yet
Developer experience	14%1 favorable, 6 not, 7 authors	Only concerns on record0 favorable, 2 not, 2 authors	27%3 favorable, 8 not, 11 authors	25%2 favorable, 6 not, 8 authors	41%7 favorable, 10 not, 17 authors	No accounts yet	Only concerns on record0 favorable, 1 not, 1 author	No accounts yet	No accounts yet	No accounts yet	25%3 favorable, 9 not, 12 authors	No accounts yet	25%2 favorable, 6 not, 8 authors	38%5 favorable, 8 not, 13 authors	38%6 favorable, 10 not, 16 authors	50%7 favorable, 7 not, 14 authors	25%3 favorable, 9 not, 12 authors	29%2 favorable, 5 not, 7 authors	25%3 favorable, 9 not, 12 authors	53%10 favorable, 9 not, 19 authors	29%4 favorable, 10 not, 14 authors	57%4 favorable, 3 not, 7 authors	45%5 favorable, 6 not, 11 authors	67%8 favorable, 4 not, 12 authors	Needs 5 accounts from 3 authors2 favorable, 2 not, 4 authors	50%3 favorable, 3 not, 6 authors	33%3 favorable, 6 not, 9 authors	39%7 favorable, 11 not, 17 authors	Only concerns on record0 favorable, 7 not, 7 authors	40%8 favorable, 12 not, 20 authors	No accounts yet	35%8 favorable, 15 not, 23 authors	No accounts yet	9%1 favorable, 10 not, 11 authors	No accounts yet	Only concerns on record0 favorable, 1 not, 1 author	13%2 favorable, 14 not, 16 authors	Only concerns on record0 favorable, 9 not, 9 authors	25%5 favorable, 15 not, 20 authors	No accounts yet	No accounts yet	Only concerns on record0 favorable, 7 not, 7 authors	11%1 favorable, 8 not, 9 authors	Only concerns on record0 favorable, 3 not, 3 authors	33%5 favorable, 10 not, 15 authors	No accounts yet	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 1 not, 1 author	No accounts yet	Only concerns on record0 favorable, 6 not, 6 authors	Only concerns on record0 favorable, 1 not, 1 author	17%1 favorable, 5 not, 6 authors	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 1 not, 1 author	No accounts yet	No accounts yet
Cost	60%3 favorable, 2 not, 5 authors	80%8 favorable, 2 not, 9 authors	74%14 favorable, 5 not, 19 authors	79%22 favorable, 6 not, 26 authors	29%7 favorable, 17 not, 23 authors	57%4 favorable, 3 not, 7 authors	Only concerns on record0 favorable, 1 not, 1 author	50%3 favorable, 3 not, 6 authors	Needs 5 accounts from 3 authors1 favorable, 3 not, 4 authors	63%5 favorable, 3 not, 8 authors	55%12 favorable, 10 not, 21 authors	Needs 5 accounts from 3 authors2 favorable, 1 not, 3 authors	69%20 favorable, 9 not, 29 authors	45%10 favorable, 12 not, 22 authors	56%15 favorable, 12 not, 27 authors	20%4 favorable, 16 not, 20 authors	43%3 favorable, 4 not, 7 authors	39%7 favorable, 11 not, 17 authors	8%1 favorable, 11 not, 12 authors	40%6 favorable, 9 not, 14 authors	50%7 favorable, 7 not, 13 authors	50%6 favorable, 6 not, 12 authors	33%7 favorable, 14 not, 20 authors	53%8 favorable, 7 not, 15 authors	29%4 favorable, 10 not, 12 authors	38%3 favorable, 5 not, 8 authors	35%7 favorable, 13 not, 20 authors	58%14 favorable, 10 not, 23 authors	Only praise on record7 favorable, 0 not, 7 authors	50%8 favorable, 8 not, 16 authors	Only concerns on record0 favorable, 1 not, 1 author	66%25 favorable, 13 not, 37 authors	Only praise on record2 favorable, 0 not, 2 authors	40%10 favorable, 15 not, 25 authors	71%5 favorable, 2 not, 7 authors	Needs 5 accounts from 3 authors1 favorable, 3 not, 4 authors	27%3 favorable, 8 not, 11 authors	Only concerns on record0 favorable, 8 not, 8 authors	19%5 favorable, 21 not, 26 authors	20%1 favorable, 4 not, 5 authors	Only praise on record4 favorable, 0 not, 4 authors	80%12 favorable, 3 not, 15 authors	50%6 favorable, 6 not, 12 authors	76%16 favorable, 5 not, 21 authors	35%8 favorable, 15 not, 23 authors	71%5 favorable, 2 not, 7 authors	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 9 not, 9 authors	Needs 5 accounts from 3 authors2 favorable, 1 not, 3 authors	Only praise on record1 favorable, 0 not, 1 author	Only praise on record1 favorable, 0 not, 1 author	Needs 5 accounts from 3 authors2 favorable, 2 not, 4 authors	Only concerns on record0 favorable, 2 not, 2 authors	Needs 5 accounts from 3 authors1 favorable, 1 not, 2 authors	Only praise on record3 favorable, 0 not, 3 authors	Only praise on record1 favorable, 0 not, 1 author	Only praise on record1 favorable, 0 not, 1 author
Speed	85%23 favorable, 4 not, 27 authors	83%5 favorable, 1 not, 5 authors	37%7 favorable, 12 not, 19 authors	45%5 favorable, 6 not, 11 authors	41%7 favorable, 10 not, 17 authors	Only praise on record1 favorable, 0 not, 1 author	Needs 5 accounts from 3 authors1 favorable, 1 not, 2 authors	80%4 favorable, 1 not, 5 authors	Needs 5 accounts from 3 authors1 favorable, 1 not, 2 authors	No accounts yet	57%4 favorable, 3 not, 7 authors	Only praise on record1 favorable, 0 not, 1 author	74%14 favorable, 5 not, 19 authors	88%14 favorable, 2 not, 16 authors	46%6 favorable, 7 not, 13 authors	44%4 favorable, 5 not, 9 authors	60%3 favorable, 2 not, 5 authors	44%4 favorable, 5 not, 9 authors	57%4 favorable, 3 not, 7 authors	31%4 favorable, 9 not, 13 authors	63%5 favorable, 3 not, 8 authors	17%1 favorable, 5 not, 6 authors	36%4 favorable, 7 not, 11 authors	73%8 favorable, 3 not, 11 authors	78%7 favorable, 2 not, 9 authors	75%6 favorable, 2 not, 8 authors	45%5 favorable, 6 not, 11 authors	61%19 favorable, 12 not, 31 authors	53%10 favorable, 9 not, 17 authors	35%6 favorable, 11 not, 16 authors	Needs 5 accounts from 3 authors2 favorable, 1 not, 3 authors	68%21 favorable, 10 not, 30 authors	Only concerns on record0 favorable, 1 not, 1 author	21%4 favorable, 15 not, 19 authors	Only concerns on record0 favorable, 2 not, 2 authors	No accounts yet	21%3 favorable, 11 not, 14 authors	Only concerns on record0 favorable, 2 not, 2 authors	29%5 favorable, 12 not, 17 authors	Needs 5 accounts from 3 authors1 favorable, 2 not, 3 authors	Only praise on record2 favorable, 0 not, 2 authors	44%4 favorable, 5 not, 9 authors	65%13 favorable, 7 not, 20 authors	33%3 favorable, 6 not, 9 authors	20%3 favorable, 12 not, 15 authors	Needs 5 accounts from 3 authors1 favorable, 1 not, 2 authors	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 1 not, 1 author	Needs 5 accounts from 3 authors1 favorable, 1 not, 2 authors	Needs 5 accounts from 3 authors3 favorable, 1 not, 4 authors	No accounts yet	33%7 favorable, 14 not, 21 authors	Needs 5 accounts from 3 authors1 favorable, 3 not, 4 authors	Only concerns on record0 favorable, 4 not, 4 authors	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 2 not, 2 authors	Only concerns on record0 favorable, 1 not, 1 author
Reliability	Only concerns on record0 favorable, 4 not, 4 authors	Only concerns on record0 favorable, 1 not, 1 author	13%1 favorable, 7 not, 8 authors	Only concerns on record0 favorable, 5 not, 5 authors	14%2 favorable, 12 not, 14 authors	Only concerns on record0 favorable, 4 not, 4 authors	Only concerns on record0 favorable, 1 not, 1 author	No accounts yet	Only concerns on record0 favorable, 2 not, 2 authors	Only concerns on record0 favorable, 4 not, 4 authors	17%1 favorable, 5 not, 6 authors	Only concerns on record0 favorable, 3 not, 3 authors	20%1 favorable, 4 not, 5 authors	18%2 favorable, 9 not, 11 authors	13%1 favorable, 7 not, 8 authors	9%1 favorable, 10 not, 11 authors	13%1 favorable, 7 not, 8 authors	63%5 favorable, 3 not, 7 authors	22%2 favorable, 7 not, 9 authors	17%2 favorable, 10 not, 12 authors	18%2 favorable, 9 not, 11 authors	17%1 favorable, 5 not, 6 authors	20%3 favorable, 12 not, 15 authors	38%6 favorable, 10 not, 16 authors	29%2 favorable, 5 not, 7 authors	40%4 favorable, 6 not, 10 authors	5%1 favorable, 18 not, 19 authors	15%3 favorable, 17 not, 20 authors	Only concerns on record0 favorable, 5 not, 5 authors	14%3 favorable, 19 not, 22 authors	Only concerns on record0 favorable, 1 not, 1 author	42%5 favorable, 7 not, 12 authors	Needs 5 accounts from 3 authors1 favorable, 1 not, 2 authors	8%1 favorable, 12 not, 13 authors	Only concerns on record0 favorable, 1 not, 1 author	Needs 5 accounts from 3 authors1 favorable, 1 not, 2 authors	38%3 favorable, 5 not, 8 authors	Only concerns on record0 favorable, 4 not, 4 authors	22%2 favorable, 7 not, 9 authors	Only concerns on record0 favorable, 1 not, 1 author	No accounts yet	60%3 favorable, 2 not, 5 authors	33%2 favorable, 4 not, 5 authors	Only concerns on record0 favorable, 7 not, 7 authors	Only concerns on record0 favorable, 4 not, 4 authors	Only concerns on record0 favorable, 1 not, 1 author	No accounts yet	Only concerns on record0 favorable, 2 not, 2 authors	Only concerns on record0 favorable, 1 not, 1 author	Needs 5 accounts from 3 authors1 favorable, 3 not, 4 authors	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 9 not, 9 authors	Only concerns on record0 favorable, 1 not, 1 author	Only concerns on record0 favorable, 2 not, 2 authors	No accounts yet	No accounts yet	Only concerns on record0 favorable, 1 not, 1 author

The 82 accounts behind gpt-oss-120b's figures

GILD searched Hacker News, GitHub, X and the web for accounts that name gpt-oss-120b.

Quality and reasoning 58%

Favorable
and they both had the correct answer
Open models by OpenAI. podnami, Hacker News commenter, self-described, Aug 5, 2025. GPT OSS 120B (gpt-oss-120b) on OpenRouter against o3 on ChatGPT.com.
Unfavorable
The 120B model badly hallucinates facts on the level of a 0.6B model.
Open models by OpenAI. dust42, Hacker News commenter, self-described, Aug 5, 2025. gpt-oss-120b asked about Mercantour national park as a hallucination test.
Favorable
IMO gpt-oss-120b is actually a very competent local coding agent
Show HN: Octofriend, a cute coding agent that can swap between GPT-5 and Claude. reissbaker, Hacker News commenter and developer of the Octo coding agent, self-described, Aug 7, 2025. gpt-oss-120b as a local coding agent while testing the Octo coding agent.
Favorable
GPT-OSS-120B Q8 was close to DeepSeek R1 671B Q16 in solving graduate-level math
GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2. storus, Hacker News commenter, self-described, Aug 10, 2025. GPT-OSS-120B at Q8 on graduate-level math against DeepSeek R1 671B.
Unfavorable
In my benchmarks oss-120 is not very smart.
GPT‑5 was the push I needed to switch to open‑source. WorldPeas, Hacker News commenter, self-described, Aug 12, 2025. gpt-oss-120b (written oss-120) in the commenter's own benchmarks.
Unfavorable
gpt-oss-120B can be dramatically moronical
DeepSeek-v3.1. mdp2021, Hacker News commenter, self-described, Aug 22, 2025. gpt-oss-120B in the commenter's own use.
Favorable
gpt-oss-120b works as well as o4-mini over my evals
DeepSeek-v3.1. petesergeant, Hacker News commenter, self-described, Aug 22, 2025. gpt-oss-120b on Cerebras through OpenRouter over the commenter's eval suite.
Favorable
gpt-oss:120b is shockingly better than what I thought it would be
Nvidia DGX Spark: When benchmark numbers meet production reality. stuckinhell, Hacker News commenter, self-described, Oct 26, 2025. gpt-oss:120b (gpt-oss-120b) in Ollama on a DGX Spark.
Favorable
The best local coding setup I've been able to do so far is using Aider with gpt-oss-120b.
Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop. lreeves, Hacker News commenter, self-described, Oct 31, 2025. gpt-oss-120b in Aider on a llama.cpp GPU server.
Favorable
we set up gpt-oss-120b on a 128GB RAM Macbook pro and it is shockingly usable
Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop. juujian, Hacker News commenter, self-described, Oct 31, 2025. gpt-oss-120b on a 128GB MacBook Pro with the Ollama UI.
Favorable
gpt-oss-120b is amazing. I created a RAG agent to hold most of GCP documentation
Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop. fm2606, Hacker News commenter, self-described, Oct 31, 2025. gpt-oss-120b in a local RAG agent over GCP documentation on an i7 with 64GB RAM and an 8GB GPU.
Favorable
GPT-OSS-120b is surprisingly good
Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop. loudmax, Hacker News commenter, self-described, Oct 31, 2025. GPT-OSS-120b heavily quantized in llama.cpp on an RTX 3090 with 128GB RAM.
Favorable
with reasoning_effort set to high, it gets it right maybe 95% of the times
A guide to local coding models. embedding-shape, Hacker News commenter, self-described, Dec 22, 2025. GPT-OSS-120B at high reasoning effort for programming in Codex.
Favorable
It's like 10x faster than gpt5-mini and yet seems to perform just as well.
How I use every Claude Code feature. riskable, Hacker News commenter, self-described, Nov 2, 2025. gpt-oss:120b (gpt-oss-120b) on Ollama cloud for coding a collaborative editor.
Favorable
the only good-enough model I still use it gpt-oss-120b-mxfp4
Cerebras Code now supports GLM 4.6 at 1000 tokens/sec. behnamoh, Hacker News commenter, self-described, Nov 8, 2025. gpt-oss-120b-mxfp4 run locally.
Favorable
It's a decent general model too
GPT-OSS 120B Runs at 3000 tokens/sec on Cerebras. drewbitt, Hacker News commenter, self-described, Nov 8, 2025. gpt-oss-120b on Cerebras through llm and Raycast since August 2025.
Unfavorable
GPT-5 and Sonnet 4.5 remain wildly more capable than gpt-oss:120b
Using Codex CLI with gpt-oss:120b on an NVIDIA DGX Spark via Tailscale. Simon Willison, independent developer and blogger, self-described, Nov 6, 2025. gpt-oss:120b (gpt-oss-120b) in Ollama on a DGX Spark driven by Codex CLI over Tailscale.
Unfavorable
vs. larger like GPT-OSS-120B at 12.5%
LLM Output Drift in Financial Workflows: Validation and Mitigation (arXiv). raffisk, Hacker News commenter posting an empirical output-consistency study, self-described, Nov 12, 2025. GPT-OSS-120B at temperature 0 on regulated financial RAG, JSON and SQL tasks.
Unfavorable
ChatGPT 4o-mini, 5 mini and OSS 120B gave me wrong answers
GPT-5.1: A smarter, more conversational ChatGPT. hamburgererror, Hacker News commenter, self-described, Nov 13, 2025. OSS 120B (gpt-oss-120b) on duck.ai.
Favorable
After switching to gpt-oss:120b it did some things quite well
Google Antigravity. onion2k, Hacker News commenter, self-described, Nov 18, 2025. gpt-oss:120b (gpt-oss-120b) in Google Antigravity.
Favorable
I've been doing the same with GPT-OSS-120B and have been impressed.
AI coding assistants are getting worse?. Eggpants, Hacker News commenter, self-described, Jan 8, 2026. GPT-OSS-120B in Claude Code.
Favorable
My local GPT-OSS 120B is fantastic at generating the scene
Games Workshop bans staff from using AI. rpdillon, Hacker News commenter, self-described, Jan 13, 2026. a local GPT-OSS 120B (gpt-oss-120b) for solo tabletop RPG scenes.
Favorable
the model found that swapping the order of the IQ imbalance coefficients fixed the performance degradation
Gas Town Decoded. spmurrayzzz, Hacker News commenter, self-described, Jan 19, 2026. gpt-oss-120b re-applying a reverted commit in a private radio-firmware repo.
Unfavorable
gpt-oss is hilariously stupid, but might work for typescript, react, python simple tasks
How to run Qwen 3.5 locally. rurban, Hacker News commenter, self-described, Mar 8, 2026. gpt-oss-120b locally with vLLM and opencode on C code.
Unfavorable
it's a noticeable regression when I hit a limit and move to kilocode with gpt-oss:120b running locally
OpenAI's latest repo has Claude as the third top contributor. datadrivenangel, Hacker News commenter, self-described, Mar 25, 2026. gpt-oss:120b (gpt-oss-120b) locally in Kilo Code.
Favorable
gpt-oss:120b scored 9.87/10, followed by gpt-5.4 at 9.65/10
Show HN: Which AI model is best for real data analysis?. pplonski86, Hacker News commenter posting his data analysis benchmark, self-described, Apr 14, 2026. gpt-oss:120b (gpt-oss-120b) on a 23-workflow data analysis benchmark.
Favorable
my results show that gpt-oss-120b (non batch via openrotuer) is the best for now for my use case
DeepSeek v4. kiproping, Hacker News commenter, self-described, Apr 24, 2026. gpt-oss-120b through OpenRouter for JSON analysis work.
Favorable
even gpt-oss-120b are solid at reasonable quants if they have the right data
Local AI needs to be the norm. tristor, Hacker News commenter, self-described, May 11, 2026. gpt-oss-120b locally at reasonable quants with search and MCP data.
Unfavorable
Haiku (cheapest Anthropic model) vs gpt-oss-120b had a big difference in quality and syntax issues
Ask HN: What coding agents are you using?. m-dot-reviews, Hacker News commenter, self-described, Jun 11, 2026. gpt-oss-120b against Claude Haiku generating code in a DSL.
Unfavorable
but not terribly smart
Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?. lambda, Hacker News commenter, self-described, Jun 15, 2026. GPT-OSS 120B (gpt-oss-120b) on a 128GB Strix Halo laptop with llama.cpp.
Favorable
The gpt-oss-120b also got it correct on my prompt version.
A global workspace in language models. AlexDragusin, Hacker News commenter, self-described, Jul 7, 2026. gpt-oss-120b on a band-identification knowledge prompt.
Favorable
gpt-oss-120b is still very useful
Cerebras CS-4. 0xbadcafebee, Hacker News commenter, self-described, Aug 19, 2026. gpt-oss-120b on Cerebras through OpenRouter.
Unfavorable
i had some pretty meh results using gpt-oss-120b it in my evals
Mercury 2.5 LLM hits 770 tokens per second. shard972, Hacker News commenter, self-described, Sep 24, 2026. gpt-oss-120b in the commenter's evals.
Unfavorable
Qwen is just the much smarter model of the 2 public options on Cerebras.
Mercury 2.5 LLM hits 770 tokens per second. scosman, Hacker News commenter, self-described, Sep 24, 2026. GPT OSS 120 (gpt-oss-120b) against Qwen on Cerebras.
Favorable
I've used it on a few for fun projects and its decent
Mercury 2.5 LLM hits 770 tokens per second. bearjaws, Hacker News commenter, self-described, Sep 23, 2026. Cerebras gpt-oss-120b on fun projects.
Unfavorable
gpt-oss-120b for example, is almost 4x the size, and performs way worse at coding tasks
Nvidia is the central bank of AI. kennywinker, Hacker News commenter, self-described, Sep 12, 2026. gpt-oss-120b on coding tasks against Qwen3.8 27B.
Unfavorable
Even GPT-OSS-120B got it wrong
Qwen 3.8 27B. dexterlagan, Hacker News commenter, self-described, Aug 15, 2026. GPT-OSS-120B on a general-knowledge question about John Keel.
Unfavorable
a 27B model (Qwen 3.6 27B) from this year is a huge improvement over a 120B model (gpt-oss:120b) from last year
Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows. xscott, Hacker News commenter, self-described, Aug 10, 2026. gpt-oss:120b (gpt-oss-120b) against Qwen 3.6 27B for programming.
Unfavorable
much worse at multilingual tasks (33 languages evaluated), and had worse instruction following
Nvidia Nemotron 3 Family of Models. kgeist, Hacker News commenter, self-described, Dec 17, 2025. gpt-oss 120b (gpt-oss-120b) on vLLM against Qwen3-Next 80b on internal production benchmarks.
Unfavorable
But it's mostly wrong.
Cerebras Deploys Trillion-Parameter AI Models. keyle, Hacker News commenter, self-described, Oct 25, 2026. GPT OSS 120b (gpt-oss-120b) on Cerebras.

Tool use and structured output 27%

Unfavorable
It repeatedly failed to use the tools correctly
Open models by OpenAI. andai, Hacker News commenter, self-described, Aug 6, 2025. gpt-oss-120b with Cerebras in Roo Code.
Unfavorable
Gpt-oss-120B breaks in some cases
Open models by OpenAI. kgeist, Hacker News commenter, self-described, Aug 6, 2025. gpt-oss-120b in llama.cpp running the commenter's tool-calling scripts, partly offloaded from the GPU.
Favorable
found most of the bigger Qwen models plus gpt-oss-120b to be great in agentic mode
How I code with AI on a budget/free. navbaker, Hacker News commenter, self-described, Aug 10, 2025. gpt-oss-120b in Continue for VS Code in agentic mode.
Unfavorable
Its tool use is quite lacking
GPT‑5 was the push I needed to switch to open‑source. WorldPeas, Hacker News commenter, self-described, Aug 12, 2025. gpt-oss-120b (written oss-120) in the commenter's own benchmarks.
Favorable
So far it works great for me with GPT-OSS-120b as a local agent, with access to search tools.
GPT-5.1: A smarter, more conversational ChatGPT. embedding-shape, Hacker News commenter, self-described, Nov 13, 2025. GPT-OSS-120b as a local agent with search tools and a verify_claim tool.
Unfavorable
Kimi is noticeably better at tool calling than gpt-oss-120b.
You should write an agent. ay, Hacker News commenter, self-described, Nov 7, 2025. gpt-oss-120b against Kimi in the commenter's toy agent that swaps turns on tool-calling mistakes.
Unfavorable
Now it's stuck in a loop repeating the last thing it output. I've seen that a lot on gpt-oss models
Google Antigravity. onion2k, Hacker News commenter, self-described, Nov 18, 2025. gpt-oss:120b (gpt-oss-120b) in Google Antigravity.
Unfavorable
proceeded to watch it run `cat README.md` over and over again
A guide to local coding models. jszymborski, Hacker News commenter, self-described, Dec 22, 2025. gpt-oss-120b in Cline on an RTX 5090.
Favorable
Once I did that it started functioning extremely well, and it's the main model I use for my homemade agents.
GLM-4.7: Advancing the Coding Capability. nekitamo, Hacker News commenter, self-described, Dec 23, 2025. GPT-oss-120B (gpt-oss-120b) in the commenter's homemade agents.
Favorable
Everything worked fine on GPT but Qwen as often as not preferred to pretend to call a tool and not actually call it.
Can I run AI locally?. adamkittelson, Hacker News commenter, self-described, Mar 13, 2026. GPT-OSS-120B (gpt-oss-120b) locally for a Discord chat bot with tool use.
Unfavorable
"openai.gpt-oss-120b-1:0" # tool calling broken
Mistral Medium 3.5. nijave, Hacker News commenter, self-described, Apr 29, 2026. openai.gpt-oss-120b on AWS Bedrock in a Strands Agents devops agent.
Unfavorable
screws up tool calls most of the time, forgets to end thinking blocks
Mercury 2.5 LLM hits 770 tokens per second. LoganDark, Hacker News commenter, self-described, Sep 24, 2026. gpt-oss-120b over Cerebras in an agent.
Unfavorable
`tool_choice="required"` is silently ignored
[Bug]: [gpt-oss-120b] tool_choice="required" ignored on /v1/chat/completions (v0.20.1). ekrako, GitHub issue author on vllm-project/vllm, self-described, Jun 1, 2026. openai/gpt-oss-120b on vLLM v0.20.1 with two H100s and the openai tool-call parser.
Unfavorable
gpt-oss:120b - thinking being returned in tool calls
gpt-oss:120b - thinking being returned in tool calls. lefoulkrod, GitHub issue author on ollama/ollama, self-described, Sep 6, 2025. gpt-oss:120b (gpt-oss-120b) on Ollama 0.11.10 in a code review agent with tools.
Unfavorable
When a chat completion request combines `tools` with `response_format: json_schema`, gpt-oss-120b never emits a tool call.
gpt-oss-120b: tools + response_format json_schema combined -> model loops in reasoning channel, never emits tool call (terminates at token budget, tool_calls=0). emirsaffar-collab, GitHub issue author on ggml-org/llama.cpp, self-described, Oct 12, 2026. gpt-oss-120b MXFP4 GGUF on llama-server, CPU inference.

Long context 25%

Unfavorable
once more than ~10k tokens you're basically queueing everything for a long time
Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs. wcallahan, Hacker News commenter, self-described, Aug 7, 2025. GPT-OSS-120B offline on a MacBook Pro M4 with 128GB RAM during a flight.
Unfavorable
Prompt processing/context however seems to be happening at about 1/4 of those models
GPT-OSS-120B runs on just 8GB VRAM & 64GB+ system RAM. tyfon, Hacker News commenter, self-described, Aug 11, 2025. gpt-oss-120b (oss-120b) through Ollama with Open WebUI, nvim and a Discord bot on 8GB VRAM and system RAM.
Unfavorable
it forces you to narrow your prompts (due to smaller context window)
How I use every Claude Code feature. riskable, Hacker News commenter, self-described, Nov 2, 2025. gpt-oss:120b (gpt-oss-120b) on Ollama cloud for coding a collaborative editor.
Unfavorable
Only gotcha is Claude code expects a 200k context window while that model max supports 130k or so. I have to do a /compress when it gets close.
AI coding assistants are getting worse?. Eggpants, Hacker News commenter, self-described, Jan 8, 2026. GPT-OSS-120B in Claude Code.
Favorable
like gpt-oss-120b, it's snappy enough even with big context
Arm's Cortex A725 Ft. Dell's Pro Max with GB10. justaboutanyone, Hacker News commenter, self-described, Jan 28, 2026. gpt-oss-120b on a Dell Pro Max with GB10.
Unfavorable
when I tried gpt-oss 120b with codex, it very easily forgets something present in the system prompt
Step 3.5 Flash – Open-source foundation model, supports deep reasoning at speed. tarruda, Hacker News commenter, self-described, Feb 19, 2026. gpt-oss 120b (gpt-oss-120b) with Codex.
Favorable
the model is super optimized and only uses 4.8 GB of additional RAM for KV cache at this context size
MacBook Pro with M5 Pro and M5 Max. super_mario, Hacker News commenter, self-described, Mar 3, 2026. gpt-oss 120b (gpt-oss-120b) on Ollama with a 128k context on an M4 Max Mac Studio.
Unfavorable
by the end, it seems to have forgotten that I had asked a question at all!
Codex with gpt-oss:120b model appears to forget the task. zhangwen0411, GitHub issue author on openai/codex, self-described, Aug 14, 2025. gpt-oss:120b (gpt-oss-120b) served by Ollama in Codex CLI 0.21.0 on a large repo.

Developer experience 14%

Favorable
running the GPT-OSS-120B on H100 using vLLM is flawless in comparison
Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs. mutkach, Hacker News commenter, self-described, Aug 7, 2025. GPT-OSS-120B on a single H100 with vLLM.
Unfavorable
a lot of CLI/TUI coding tools (e.g., opencode) were not working reliably offline at this time with the model
Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs. wcallahan, Hacker News commenter, self-described, Aug 7, 2025. GPT-OSS-120B offline on a MacBook Pro M4 with 128GB RAM during a flight.
Unfavorable
that still required me to manually hack in support for changing the temperature, and changing the prompts Codex use a bit, to get OK results
GLM 4.5 with Claude Code. diggan, Hacker News commenter, self-described, Sep 6, 2025. GPT-OSS-120b locally with Codex CLI.
Unfavorable
I needed to do some local patches to Codex as they don't allow configuring and setting the right values for temperature and top_p for GPT-OSS
Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop. embedding-shape, Hacker News commenter, self-described, Oct 31, 2025. GPT-OSS-120b at high reasoning effort with Codex and llama.cpp, all local.
Unfavorable
GPT-oss-120B was also completely failing for me, until someone on reddit pointed out that you need to pass back in the reasoning tokens
GLM-4.7: Advancing the Coding Capability. nekitamo, Hacker News commenter, self-described, Dec 23, 2025. GPT-oss-120B (gpt-oss-120b) in the commenter's homemade agents.
Unfavorable
it feels less like a science experiment than any previous local model I've run, including GPT OSS 120B
Running local models on an M4 with 24GB memory. soganess, Hacker News commenter, self-described, May 11, 2026. GPT OSS 120B (gpt-oss-120b) run locally on an M5 Max, compared with Gemma 4 31B.
Unfavorable
Harmony: bad request on gpt-oss-120b and tool calls with `create_react_agent`
Harmony: bad request on gpt-oss-120b and tool calls with create_react_agent. d-r-e, GitHub issue author on langchain-ai/langchain, self-described, Sep 8, 2025. openai/gpt-oss-120b through langchain-openai's Responses API in a LangGraph ReAct agent with one tool.

Cost 60%

Favorable
like 1/5th the price of o4-mini
DeepSeek-v3.1. petesergeant, Hacker News commenter, self-described, Aug 22, 2025. gpt-oss-120b on Cerebras through OpenRouter over the commenter's eval suite.
Favorable
if we look at price/performance, GPT-OSS is quite impressive
A first look at GPT-OSS-120B's coding ability. Jonathan Ellis, author for Brokk's Power Ranking coding benchmark, self-described, Aug 5, 2025. GPT-OSS-120B (gpt-oss-120b) on the Brokk Power Ranking Java coding benchmark.
Favorable
Gpt-oss 120b is much more focused and steerable there.
Mistral 3 family of models released. data-ottawa, Hacker News commenter, self-described, Dec 2, 2025. Gpt-oss 120b (gpt-oss-120b) against Qwen3 models on thinking-token use.
Favorable
Oh dang I had no idea that gpt-oss-120b was that cheap these days.
Pi – A minimal terminal coding harness. h14h, Hacker News commenter, self-described, Feb 26, 2026. gpt-oss-120b on Vultr's inference service as a reasoning tool for a Kimi agent.
Unfavorable
Think: $5–10/minute levels of cost with a single agent, because Cerebras also offers no cache pricing for input tokens at all.
Mercury 2.5 LLM hits 770 tokens per second. LoganDark, Hacker News commenter, self-described, Sep 24, 2026. gpt-oss-120b over Cerebras in an agent.
Unfavorable
very prone to overthinking, so individual turns may be 3-10k tokens of just thinking + input + output
Qwen 3.8 27B available on Cerebras at 1500 tokens/s. nostrebored, Hacker News commenter, self-described, Sep 3, 2026. gpt-oss-120b on Cerebras for coding tasks.

Speed 85%

Favorable
on OpenRouter GPT OSS 120B thought for 0.7 seconds
Open models by OpenAI. podnami, Hacker News commenter, self-described, Aug 5, 2025. GPT OSS 120B (gpt-oss-120b) on OpenRouter against o3 on ChatGPT.com.
Unfavorable
The latency also seemed pretty high
Open models by OpenAI. andai, Hacker News commenter, self-described, Aug 6, 2025. gpt-oss-120b with Cerebras in Roo Code.
Favorable
the throughput stays at 130-140 t/s for a single H100
Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs. mutkach, Hacker News commenter, self-described, Aug 7, 2025. GPT-OSS-120B on a single H100 with vLLM.
Favorable
much faster with way fewer thinking tokens
GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2. storus, Hacker News commenter, self-described, Aug 10, 2025. GPT-OSS-120B at Q8 on graduate-level math against DeepSeek R1 671B.
Favorable
I can't really notice any difference between the speed of oss-120b once the context is processed and claude opus-4 via api
GPT-OSS-120B runs on just 8GB VRAM & 64GB+ system RAM. tyfon, Hacker News commenter, self-described, Aug 11, 2025. gpt-oss-120b (oss-120b) through Ollama with Open WebUI, nvim and a Discord bot on 8GB VRAM and system RAM.
Favorable
I am paying Ollama $20/month just to run gpt-oss-120b very fast
Claude Code is all you need. mark_l_watson, Hacker News commenter, self-described, Aug 12, 2025. gpt-oss-120b on Ollama's $20 a month hosted service with web search.
Favorable
I can run it via OpenRouter on Cerebras where it's SO DAMN FAST
DeepSeek-v3.1. petesergeant, Hacker News commenter, self-described, Aug 22, 2025. gpt-oss-120b on Cerebras through OpenRouter over the commenter's eval suite.
Favorable
GPT OSS has particularly good latency
Experimenting with Local LLMs on macOS. ai-christianson, Hacker News commenter, self-described, Sep 8, 2025. gpt-oss-120b on a 128GB M4 MacBook.
Favorable
I was made to believe this would be unbearably slow, but it's really not
Claude for Excel. Wowfunhappy, Hacker News commenter, self-described, Oct 28, 2025. gpt-oss 120b (gpt-oss-120b) in Ollama with about 80% of the model on the CPU next to an RTX 4080.
Unfavorable
gpt-oss-120b took over an hour but got 47 / 50
Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop. fm2606, Hacker News commenter, self-described, Oct 31, 2025. gpt-oss-120b in a local RAG agent over GCP documentation on an i7 with 64GB RAM and an 8GB GPU.
Unfavorable
too quantized and too slow to be more than a toy
Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop. loudmax, Hacker News commenter, self-described, Oct 31, 2025. GPT-OSS-120b heavily quantized in llama.cpp on an RTX 3090 with 128GB RAM.
Favorable
Ollama cloud for gpt-oss:120b (it's so fast!)
How I use every Claude Code feature. riskable, Hacker News commenter, self-described, Nov 2, 2025. gpt-oss:120b (gpt-oss-120b) on Ollama cloud for coding a collaborative editor.
Favorable
I have it plugged up in llm and raycast since August at great speeds
GPT-OSS 120B Runs at 3000 tokens/sec on Cerebras. drewbitt, Hacker News commenter, self-described, Nov 8, 2025. gpt-oss-120b on Cerebras through llm and Raycast since August 2025.
Favorable
Cerebras runs the new model at between 2 and 4 thousands tokens per second!
OpenAI's new open weight (Apache 2) models are really good. Simon Willison, independent developer and blogger, self-described, Aug 5, 2025. gpt-oss-120b on Cerebras through the llm-cerebras plugin.
Favorable
Almost imperceptible transcription and processing delay.
Ask HN: How can I get better at using AI for programming?. quinncom, Hacker News commenter, self-described, Dec 13, 2025. Cerebras gpt-oss-120b for post-processing local Parakeet transcriptions in Spokenly.
Favorable
it input processed at 1650 tok/s and output generated at 27 tok/s. Just fast enough IMO
Arm's Cortex A725 Ft. Dell's Pro Max with GB10. justaboutanyone, Hacker News commenter, self-described, Jan 29, 2026. gpt-oss-120b at 128k context summarizing a 97k-token novel on a GB10.
Favorable
using GPT-OSS 120b in ollama. It's decently fast
Ask HN: Anyone Using a Mac Studio for Local AI/LLM?. caterama, Hacker News commenter, self-described, Feb 6, 2026. GPT-OSS 120b (gpt-oss-120b) in Ollama on an M3 Ultra with 256 GB.
Favorable
OSS 120B on Groq/Cerebras is one of the fastest models available with enough smarts for basic tasks
Switch to Claude without starting over. pants2, Hacker News commenter, self-described, Mar 1, 2026. OSS 120B (gpt-oss-120b) on Groq and Cerebras in production.
Favorable
easily runs inference north of 50 t/s depending on context
Can I run AI locally?. bityard, Hacker News commenter, self-described, Mar 13, 2026. gpt-oss-120b on a 128GB Strix Halo machine.
Favorable
They easily push ~400 input and ~100 output tokens per second per device on say gpt-oss-120b.
AI should elevate your thinking, not replace it. StrauXX, Hacker News commenter, self-described, Apr 26, 2026. gpt-oss-120b on clusters of DGX Spark machines for company RAG users.
Favorable
GPT-OSS 120B if I want super fast
Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?. lambda, Hacker News commenter, self-described, Jun 15, 2026. GPT-OSS 120B (gpt-oss-120b) on a 128GB Strix Halo laptop with llama.cpp.
Favorable
I was able to run the gpt-oss 120b model with pretty decent performance
Show HN: Getting GLM 5.2 running on my slow computer. efficax, Hacker News commenter, self-described, Jul 10, 2026. gpt-oss 120b (gpt-oss-120b) on a CPU workstation with a 12GB GPU.
Favorable
The speed is amazing
Mercury 2.5 LLM hits 770 tokens per second. LoganDark, Hacker News commenter, self-described, Sep 24, 2026. gpt-oss-120b over Cerebras in an agent.
Favorable
the speed is crazy to watch
Mercury 2.5 LLM hits 770 tokens per second. bearjaws, Hacker News commenter, self-described, Sep 23, 2026. Cerebras gpt-oss-120b on fun projects.
Favorable
It was really great in terms of latency/throughput
Cerebras CS-4. preommr, Hacker News commenter, self-described, Aug 19, 2026. GPT-OSS 120B (gpt-oss-120b) for a design tool.
Unfavorable
gpt-oss was slightly slower (vLLM, both fit in VRAM)
Nvidia Nemotron 3 Family of Models. kgeist, Hacker News commenter, self-described, Dec 17, 2025. gpt-oss 120b (gpt-oss-120b) on vLLM against Qwen3-Next 80b on internal production benchmarks.
Favorable
I have tried GPT OSS 120b on cerebras and it flies.
Cerebras Deploys Trillion-Parameter AI Models. keyle, Hacker News commenter, self-described, Oct 25, 2026. GPT OSS 120b (gpt-oss-120b) on Cerebras.

Reliability Only concerns on record

Unfavorable
and then I got 429 too many requests
Open models by OpenAI. andai, Hacker News commenter, self-described, Aug 6, 2025. gpt-oss-120b with Cerebras in Roo Code.
Unfavorable
makes the system somewhat unstable. Have to reboot frequently
Ask HN: Anyone Using a Mac Studio for Local AI/LLM?. caterama, Hacker News commenter, self-described, Feb 6, 2026. GPT-OSS 120b (gpt-oss-120b) in Ollama on an M3 Ultra with 256 GB.
Unfavorable
requests intermittently fail with HTTP 500 due to `openai_harmony.HarmonyError`
[Bug]: openai_harmony.HarmonyError: unexpected tokens remaining in message header on v0.26.0 (gpt-oss-120b). JakobKe, GitHub issue author on vllm-project/vllm, self-described, Aug 12, 2026. openai/gpt-oss-120b on vLLM v0.26.0 with tool calling on an H100.
Unfavorable
GPT-OSS 120B not working with 0.30
GPT-OSS 120B not working with 0.30. maurizioaiello, GitHub issue author on ollama/ollama, self-described, Jun 2, 2026. GPT-OSS 120B (gpt-oss-120b) on Ollama 0.30 on an AMD Strix Halo with Open WebUI.