What teams building AI features on LiteLLM report, with every finding tied to its evidence.
Read the findingsPractitioner sentiment, teams and leaders
+8 Mixed from 37 sources that take a side
How this figure is madeFrom 36 authors in 33 discussions. Run a pilot on one service's model traffic to measure added latency, failed fallbacks and how closely its spend records match each provider's bill.
Jan 2025 to Oct 2026.
125 sources from practitioners, by month and sentiment.
Teams and leaders are mixed, and most of the mixed accounts are bug reports. Okta and Zalando run LiteLLM as the company gateway, and teams on Reddit cap each team key with a daily hard stop. Critics report budgets that track spend but do not block requests, and team keys that get messy past a handful of teams. Every team and leadership source that discusses stability raises it. On performance at scale, 7 of 9 raise concerns, and 5 of the 9 who describe a gateway choice picked another tool or their own code.
Positive, mixed and negative sources
The score counts them as 9 positive, 21 mixed and 6 negative votes, one per author. See How the numbers work.
See the evidenceBelow: 4 praise themes, 3 concerns, 2 close to agreement and 2 disagreements.
Counted for Teams and technical leaders, from 47 sources. Change group
Every capability practitioners describe positively, ranked by how many sources say so. Open one to read who said it.
Counted for Teams and technical leaders.
Positive, 7
Positive
9dev, Hacker News commenter
Comment on: Litelm: LiteLLM Without the BloatHacker News comment
Deploying LiteLLM in their platform for its reliable per-customer token spend tracking across services, the features slimmer forks remove.
Positive
kgeist, Hacker News commenter
Comment on: LLMCap - A proxy that hard-stops LLM API calls when you hit a dollar capHacker News comment
Their team self-hosts LiteLLM for free and sets budgets per day or week.
Positive
intothemild, Hacker News commenter
Comment on: AI subscriptions are a ticking time bomb for enterpriseHacker News comment
Put self-hosted Qwen, Gemma, and DeepSeek behind LiteLLM to watch token use and bring inference cost in-house, and the company is rolling it out more widely.
Mixed
Alina Dallmann, Speaker at PyData Berlin 2025
One API to Rule Them All? LiteLLM in ProductionConference talk, PyData Berlin 2025 (PyData YouTube channel)
Ran LiteLLM's proxy for a university chatbot with cost tracking, budgets and load balancing, but hit custom-budget limits, slow new-model support and bugs in new features.
"…deploying the the proxy server we already get an admin user interface delivered as well. Um this is a screenshot um of one example of it. Um here we can do most of the process I just described earlier to assign budgets…"
Mixed
Bartosz Ocytko, Executive Principal Engineer, Zalando
Agentic Engineering at Zalando: a snapshotEngineering blog post, Zalando Engineering Blog
Zalando likes LiteLLM for its extensibility, using call hooks for anonymized cost tracking and client version checks. Stability and memory leaks force a restart every 20,000 requests, and the proxy serves 2,000 monthly users on six small pods.
"…our ML platform team deployed in January 2024 a LiteLLM based API proxy with access to models from different providers (now: OpenAI, AWS Bedrock, and Google Vertex)."
Positive
lbreakjai, At a company that rolled out LiteLLM (as stated)
Comment on: Kimi K2.7 Code is generally available in GitHub CopilotHacker News comment
Their company canceled everyone's individual AI plans and rolled LiteLLM out internally instead.
Positive
Thoughtworks delivery teams, Technology Radar
LiteLLM (Technology Radar, Languages & Frameworks)Technology Radar entry, Thoughtworks
Thoughtworks teams increasingly adopt LiteLLM as a sensible default gateway for retries, failover, load balancing, and budgets. Provider-specific features still need provider-specific parameters, and drop_params silently discards unsupported ones.
Mixed or negative, 3
Mixed
Jack Wildman, Senior Software Engineer, FutureSearch
LiteLLM made us accidentally make our product free for a weekEngineering blog post, FutureSearch blog
A LiteLLM upgrade began stripping caller tags without an error, so FutureSearch billed no usage for 6.5 days. It calls the default-on change a footgun, added alerts, and stayed on LiteLLM at v1.86.4.
"A routine LiteLLM version bump turned off our usage billing for six and a half days. We had many alerts for runaway spend, but not for spend dropping to zero."
Mixed
lin0703, GitHub user
[Bug]: Budget enforcement bypassed in v1.82.3 for key/user max_budget despite spend exceeding max_budgetGitHub issue
On a fresh v1.82.3 deployment, key and user max_budget were stored and spend tracked, but requests were not blocked once spend passed the budget.
Negative
smcleod, Hacker News commenter
Comment on: Control LLM Spend and Access with any-LLM-gatewayHacker News comment
After deploying LiteLLM Proxy in several places, plans to swap it for Bifrost: budget limits not enforced, parameter bugs, latency, and slow responses to reported fixes.
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.Every LiteLLM budget is enforced against spend read from the database, so none caps anything on a deployment without one. Budget reservation is on by default; fail-closed budget and rate-limit enforcement is opt-in. Budgets and rate limits
Open them in the recordCounted for Teams and technical leaders.
Mixed
Bartosz Ocytko, Executive Principal Engineer, Zalando
Agentic Engineering at Zalando: a snapshotEngineering blog post, Zalando Engineering Blog
Zalando likes LiteLLM for its extensibility, using call hooks for anonymized cost tracking and client version checks. Stability and memory leaks force a restart every 20,000 requests, and the proxy serves 2,000 monthly users on six small pods.
"…our ML platform team deployed in January 2024 a LiteLLM based API proxy with access to models from different providers (now: OpenAI, AWS Bedrock, and Google Vertex)."
Positive
NetworkChuck (Chuck Keith), creator of the NetworkChuck YouTube channel
I'm changing how I use AI (Open WebUI + LiteLLM)Video, YouTube
Uses LiteLLM behind Open WebUI to reach ChatGPT, Claude, Gemini and Grok through one self-hosted gateway shared with a team and family under their own controls.
Positive
kgeist, Hacker News commenter
Comment on: LLMCap - A proxy that hard-stops LLM API calls when you hit a dollar capHacker News comment
Their team self-hosts LiteLLM for free and sets budgets per day or week.
Positive
intothemild, Hacker News commenter
Comment on: AI subscriptions are a ticking time bomb for enterpriseHacker News comment
Put self-hosted Qwen, Gemma, and DeepSeek behind LiteLLM to watch token use and bring inference cost in-house, and the company is rolling it out more widely.
Mixed
kingreflex, Hacker News commenter
Comment on: Malicious litellm_init.pth in litellm 1.82.8 PyPI package - credential stealerHacker News comment
During the March 2026 PyPI compromise, asked whether the team's Helm-deployed LiteLLM images on pinned stable tags were affected.
Counted for Teams and technical leaders.
Positive, 4
Mixed
Arseny Zinchenko, DevOps engineer and author of the RTFM blog
LiteLLM: OpenRouter Integration and Fallbacks ConfigurationEngineering blog post, RTFM
Added OpenRouter behind the team's existing LiteLLM gateway so services changed nothing. OpenRouter returned 429 errors in test production, so the team configured fallbacks to OpenAI in LiteLLM, keeping failover out of client code.
Positive
Thoughtworks delivery teams, Technology Radar
LiteLLM (Technology Radar, Languages & Frameworks)Technology Radar entry, Thoughtworks
Thoughtworks teams increasingly adopt LiteLLM as a sensible default gateway for retries, failover, load balancing, and budgets. Provider-specific features still need provider-specific parameters, and drop_params silently discards unsupported ones.
Mixed
sacrelege, Running a small inference provider (as stated)
Comment on: N-Day-Bench - Can LLMs find real vulnerabilities in real codebases?Hacker News comment
Their new inference service uses LiteLLM as the load balancer in front of its GPU servers, which they are replacing with a custom Rust implementation.
Mixed
Alina Dallmann, Speaker at PyData Berlin 2025
One API to Rule Them All? LiteLLM in ProductionConference talk, PyData Berlin 2025 (PyData YouTube channel)
Ran LiteLLM's proxy for a university chatbot with cost tracking, budgets and load balancing, but hit custom-budget limits, slow new-model support and bugs in new features.
"…deploying the the proxy server we already get an admin user interface delivered as well. Um this is a screenshot um of one example of it. Um here we can do most of the process I just described earlier to assign budgets…"
Mixed or negative, 1
Mixed
bjornjee, ML Ops team member (as stated)
[Bug]: Litellm Proxy availability concerns. Cascading failures during high load to providersGitHub issue
Under sustained provider 429s, new LiteLLM pods failed Kubernetes readiness probes, so all pods restarted and the gateway went down.
Counted for Teams and technical leaders.
Positive, 4
Positive
jbellis, Hacker News commenter, building Brokk
Comment on: Hypermode Model Router Preview - OpenRouter AlternativeHacker News comment
Brokk uses LiteLLM as one API over several models, but finds models not interchangeable for code, so routing to the fastest model does not appeal.
Positive
NetworkChuck (Chuck Keith), creator of the NetworkChuck YouTube channel
I'm changing how I use AI (Open WebUI + LiteLLM)Video, YouTube
Uses LiteLLM behind Open WebUI to reach ChatGPT, Claude, Gemini and Grok through one self-hosted gateway shared with a team and family under their own controls.
Mixed
Callum McMahon, Research Scientist at FutureSearch
litellm 1.82.8 Supply Chain Attack on PyPI (March 2026)Engineering blog post, FutureSearch blog
FutureSearch, which uses LiteLLM to balance quality, speed and cost across providers, caught the 1.82.8 credential stealer when an MCP plugin pulled it in; damage was minimal.
"We discovered it when the package was pulled in as a transitive dependency by an MCP plugin running inside Cursor."
Positive
bitdribble, Founder of DocRouter (as stated)
Comment on: Launch HN: Reducto Studio (YC W24) - Build accurate document pipelines, fastHacker News comment
Their document-processing startup runs on top of LiteLLM as a portability layer so it supports all major LLMs.
Mixed or negative, 3
Positive
Thoughtworks delivery teams, Technology Radar
LiteLLM (Technology Radar, Languages & Frameworks)Technology Radar entry, Thoughtworks
Thoughtworks teams increasingly adopt LiteLLM as a sensible default gateway for retries, failover, load balancing, and budgets. Provider-specific features still need provider-specific parameters, and drop_params silently discards unsupported ones.
Mixed
nicovank, Hacker News commenter, co-author of the project
Comment on: AI Meets WinDBGHacker News comment
Routes the project through LiteLLM, so other providers should work in theory, but in their experience only OpenAI's function calling works well.
Mixed
jbellis, Building the Brokk coding tool (as stated)
Comment on: The unreasonable effectiveness of an LLM agent loop with tool useHacker News comment
LiteLLM still does not support parallel tool calls with Gemini models, so the Brokk team wrote its own on top of structured output.
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.BerriAI's site lists 140+ providers and 1,892 models behind one OpenAI-compatible API; its documentation says 100+ providers. LiteLLM AI Gateway home page
Open them in the recordEvery concern practitioners raise, most raised first.
Counted for Teams and technical leaders.
Mixed
Bartosz Ocytko, Executive Principal Engineer, Zalando
Agentic Engineering at Zalando: a snapshotEngineering blog post, Zalando Engineering Blog
Zalando likes LiteLLM for its extensibility, using call hooks for anonymized cost tracking and client version checks. Stability and memory leaks force a restart every 20,000 requests, and the proxy serves 2,000 monthly users on six small pods.
"…our ML platform team deployed in January 2024 a LiteLLM based API proxy with access to models from different providers (now: OpenAI, AWS Bedrock, and Google Vertex)."
Negative
yryzhan-vitech, GitHub user running the Noma guardrail on a LiteLLM proxy
[Bug]: Noma v2 guardrail ignores streaming_end_of_stream_only and streaming_sampling_rateGitHub issue, BerriAI/litellm
Two streaming settings for the Noma guardrail had no effect: post-call scans ran per chunk at the hard-coded default sampling rate of 5 whatever the config said, with no error or warning, so the misconfiguration stayed invisible.
"Setting either in config.yaml has no effect: streaming post-call scans always run per chunk at the hard-coded default sampling rate of 5. There is no error and no warning, so the misconfiguration is invisible."
Mixed
Jack Wildman, Senior Software Engineer, FutureSearch
LiteLLM made us accidentally make our product free for a weekEngineering blog post, FutureSearch blog
A LiteLLM upgrade began stripping caller tags without an error, so FutureSearch billed no usage for 6.5 days. It calls the default-on change a footgun, added alerts, and stayed on LiteLLM at v1.86.4.
"A routine LiteLLM version bump turned off our usage billing for six and a half days. We had many alerts for runaway spend, but not for spend dropping to zero."
Mixed
ruettiger, GitHub user running LiteLLM on GKE
Sync convert_url_to_base64() blocks asyncio event loop, causing pod health check failures and mass restartsGitHub issue
In a 16-replica GKE production deployment, one unreachable image URL blocked LiteLLM's event loop for minutes, failed health checks, and left only 1 pod available.
Mixed
phanisarman, GitHub user
[Bug]: Significant performance regression after upgrading from 1.80.5 to 1.81.x (UI + API slowness)GitHub issue
Upgrading from 1.80.5 to 1.81.x made the UI and API consistently slower on the same infrastructure for a 50-user, 200-model gateway.
Mixed
Alina Dallmann, Speaker at PyData Berlin 2025
One API to Rule Them All? LiteLLM in ProductionConference talk, PyData Berlin 2025 (PyData YouTube channel)
Ran LiteLLM's proxy for a university chatbot with cost tracking, budgets and load balancing, but hit custom-budget limits, slow new-model support and bugs in new features.
"…deploying the the proxy server we already get an admin user interface delivered as well. Um this is a screenshot um of one example of it. Um here we can do most of the process I just described earlier to assign budgets…"
Negative
sothatsit, Hacker News commenter
Comment on: Control LLM Spend and Access with any-LLM-gatewayHacker News comment
Their team calls LiteLLM a dumpster fire of enterprise features and bugs: key budget edits and old-log cleanup gated as Enterprise, and a UI that flooded itself with retries.
Negative
smcleod, Hacker News commenter
Comment on: Control LLM Spend and Access with any-LLM-gatewayHacker News comment
After deploying LiteLLM Proxy in several places, plans to swap it for Bifrost: budget limits not enforced, parameter bugs, latency, and slow responses to reported fixes.
Mixed
Sanady, GitHub user at a firm running LiteLLM on Kubernetes
Heavy RAM usage over some timeGitHub issue
A firm running LiteLLM on two 4 GB Kubernetes pods sees memory fill within a day or two until the pods crash, after earlier fixes for the same leak.
Mixed
lin0703, GitHub user
[Bug]: Budget enforcement bypassed in v1.82.3 for key/user max_budget despite spend exceeding max_budgetGitHub issue
On a fresh v1.82.3 deployment, key and user max_budget were stored and spend tracked, but requests were not blocked once spend passed the budget.
Mixed
TSJasonH, GitHub user
[Bug]: Heavy RAM Usage over timeGitHub issue
Their LiteLLM container's memory grows over days until monitoring alerts fire and a restart is needed, across stable releases.
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.LiteLLM ships a stable minor release about every week and supports only the four most recent stable minor lines. A deployment more than four weeks behind receives no patches. Version support
Open them in the recordCounted for Teams and technical leaders.
Raised as a concern, 7
Mixed
Sanady, GitHub user at a firm running LiteLLM on Kubernetes
Heavy RAM usage over some timeGitHub issue
A firm running LiteLLM on two 4 GB Kubernetes pods sees memory fill within a day or two until the pods crash, after earlier fixes for the same leak.
Mixed
ruettiger, GitHub user running LiteLLM on GKE
Sync convert_url_to_base64() blocks asyncio event loop, causing pod health check failures and mass restartsGitHub issue
In a 16-replica GKE production deployment, one unreachable image URL blocked LiteLLM's event loop for minutes, failed health checks, and left only 1 pod available.
Mixed
jklinkenberg, GitHub user
[Performance] LiteLLM Overhead and Performance in ProductionGitHub issue
Benchmarked GPT-OSS-120B: about 16 requests per second direct versus about 9 requests per second through the LiteLLM proxy at 500 concurrent requests, even with spend logs disabled.
Mixed
phanisarman, GitHub user
[Bug]: Significant performance regression after upgrading from 1.80.5 to 1.81.x (UI + API slowness)GitHub issue
Upgrading from 1.80.5 to 1.81.x made the UI and API consistently slower on the same infrastructure for a 50-user, 200-model gateway.
Mixed
TSJasonH, GitHub user
[Bug]: Heavy RAM Usage over timeGitHub issue
Their LiteLLM container's memory grows over days until monitoring alerts fire and a restart is needed, across stable releases.
Negative
smcleod, Hacker News commenter
Comment on: Control LLM Spend and Access with any-LLM-gatewayHacker News comment
After deploying LiteLLM Proxy in several places, plans to swap it for Bifrost: budget limits not enforced, parameter bugs, latency, and slow responses to reported fixes.
Mixed
bjornjee, ML Ops team member (as stated)
[Bug]: Litellm Proxy availability concerns. Cascading failures during high load to providersGitHub issue
Under sustained provider 429s, new LiteLLM pods failed Kubernetes readiness probes, so all pods restarted and the gateway went down.
Discussed without concern, 2
Mixed
phenethol, LLM infrastructure intern at the Ruđer Bošković Institute
litellm-ops: operations toolkit and proxy-versus-direct latency benchmarkIndependent benchmark, GitHub repository
Built against the Ruđer Bošković Institute's LiteLLM proxy, the toolkit's benchmark finds the proxy adds about 30 ms and is no bottleneck up to 8 concurrent requests. In places, the admin UI does not match what the API does.
Positive
mpyne, On a team that adopted LiteLLM (as stated)
Comment on: Litelm: LiteLLM Without the BloatHacker News comment
Their organization chose LiteLLM after internal discussion; its added latency has not been noticeable to users and it has behaved as the team expected.
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.BerriAI is porting request translation to a Rust core, opt-in per model and in beta. Python still runs authentication, routing, logging, and spend tracking, and the standalone Rust server has no published image. Rust AI Gateway (beta)
Open them in the recordCounted for Teams and technical leaders.
Raised as a concern, 5
Mixed
Matteo Dora and David Berenstein, Giskard
LiteLLM supply chain compromise: what happened, impact on Giskard, and lessons learnedEngineering blog post, Giskard blog
Giskard's own services were safe because LiteLLM was pinned, but its open-source packages could pull in the bad releases. It traces the compromise to LiteLLM's CI pipeline, which gave Trivy every secret and published with a long-lived PyPI token.
Negative
Daniel Hnyk, FutureSearch
LiteLLM Hack: Were You One of the 47,000?Engineering blog post, FutureSearch blog
FutureSearch counted 46,996 downloads of the two compromised LiteLLM releases in the 46 minutes they were live and found 88% of the 2,337 dependent packages allowed them.
Negative
santiagobasulto, Leading a team of developers (as stated)
Comment on: Tell HN: Litellm 1.82.7 and 1.82.8 on PyPI are compromisedHacker News comment
Advised a team of less senior developers against LiteLLM and LangChain, citing supply-chain and vulnerability risk, and to call provider APIs directly.
Mixed
ddp26, Leading the FutureSearch team (as stated)
Comment on: Tell HN: Litellm 1.82.7 and 1.82.8 on PyPI are compromisedHacker News comment
Their FutureSearch team was first hit by the compromised release: uvx pulled the unpinned latest LiteLLM and Cursor started the local MCP server automatically.
Mixed
Callum McMahon, Research Scientist at FutureSearch
litellm 1.82.8 Supply Chain Attack on PyPI (March 2026)Engineering blog post, FutureSearch blog
FutureSearch, which uses LiteLLM to balance quality, speed and cost across providers, caught the 1.82.8 credential stealer when an MCP plugin pulled it in; damage was minimal.
"We discovered it when the package was pulled in as a transitive dependency by an MCP plugin running inside Cursor."
Discussed without concern, 1
Mixed
kingreflex, Hacker News commenter
Comment on: Malicious litellm_init.pth in litellm 1.82.8 PyPI package - credential stealerHacker News comment
During the March 2026 PyPI compromise, asked whether the team's Helm-deployed LiteLLM images on pinned stable tags were affected.
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.From v1.83.0, every LiteLLM Docker image on GHCR is signed with cosign, and BerriAI documents how to reject unsigned images in Kubernetes. Docker image security
Open them in the recordStatements this group's sources agree on, then the ones close to agreement.
Close to agreement: each needs 10 supporting sources
An organization needs one gateway that issues keys, sets limits, and logs every model call; provider keys stay out of individual apps.
9 supporting sourcesA share of teams and developers move to OpenRouter, Portkey, Bifrost, or a small gateway of their own.
5 supporting sourcesQuestions the sources split on, with the count on each side.
Evidence remains split
Chose or kept LiteLLM, 4 sources
Some chose LiteLLM for self-hosting with no per-token fee, spend tracking per key, and provider coverage.
Chose another tool, 5 sources
Others moved to or picked OpenRouter, Portkey, Bifrost, or their own code, most often over stability or latency.
Evidence remains split
Runs it in production, 5 sources
Some run it as their organization's gateway and report it behaving as planned, with latency their users do not notice.
Not production-grade, 3 sources
Others judge it not production-grade, citing bugs, unenforced budgets, latency, and outages.
The three groups, then every practitioner by the context they work in, and how emphasis changes with scale.
Three groups drawn from the same sources. Choose a group to select whose experience counts. Within that group, each author counts once in the practitioner sentiment, with their latest view. Signals open on Teams and technical leaders.
47 sources. Shared-gateway teams and leaders describe why they chose or rejected LiteLLM, with accounts of spend control and operations work.
78 sources. Developers describe LiteLLM in personal projects and coding tools, including self-hosted model setups.
125 sources. Together, these accounts combine platform and small product teams with solo developers. Their traffic and hosting requirements differ.
125 sources from practitioners, by the context they work in. Filled rows count for this group.
Emphasizes Coding tools behind one endpoint, stability, and performance
Emphasizes Budgets, upgrade regressions, and whether to keep it
Emphasizes Per-developer cost tracking and incident response
Emphasizes Gateway choice and supply-chain risk
Emphasizes An institute's own proxy with per-user budgets and rate limits
What changes with scale. Individual developers praise the single endpoint for coding tools and spend tracking. Stability and performance lead their concerns, and after the October 2026 compromise several moved to Bifrost. Teams and leaders run it as the company gateway at Okta and Zalando and discuss budgets, upgrade regressions and supply-chain exposure. Those who left cite bugs, unenforced budgets or security lapses, along with excessive operations work.
Where this group's sources were published.
Counted for Teams and technical leaders, from 47 sources. Change group
59 sources for this group, 12 of them documentation
Earlier evidence remains when it still applies. Records tied to changed or fixed features carry Needs a recheck.
See sources by originWho each source is, how it knows, whether it still applies and what it counts toward.
Newest first, on its own page. Every count above opens its sources there.
Read all 137 sourcesAbout this read
A running review of what teams building AI features on LiteLLM report, from individual builders to company-wide platforms. Each finding links to its sources, and each conflict shows both sides.
Counts cover only the sources GILD reviewed.
Choose a group to select whose experience counts. Within that group, each author counts once in the practitioner sentiment, with their latest view. GILD files each source in a group from how the account describes its author. A role or employer marked self-described is what the source states about themselves. Documentation checks facts for every group and never counts toward sentiment.
A source's date and whether it still applies are separate: a recent post can describe an old version, and an old one can still hold.
Flagged for a recheck, of the sources that take a side: teams and leaders 4 of 37, individual builders 5 of 71, all practitioners 9 of 108.
Each row opens its sources for the group you are reading.
| Finding | Teams and technical leaders | Individual builders | All practitioners | Sources |
|---|---|---|---|---|
| Current readCurrent sentiment | 10 positive, 21 mixed, 6 negative | 28 positive, 21 mixed, 22 negative | 38 positive, 42 mixed, 28 negative | Sources |
| PraiseOne API Across Providers | 4 of 7 positive | 8 of 17 positive | 12 of 24 positive | Sources |
| PraiseVirtual Keys, Budgets & Spend Tracking | 7 of 10 positive | 12 of 14 positive | 19 of 24 positive | Sources |
| PraiseRouting, Fallbacks & Load Balancing | 4 of 5 positive | 4 of 5 positive | 8 of 10 positive | Sources |
| PraiseSelf-Hosted Control | 5 of 5 positive | 7 of 7 positive | 12 of 12 positive | Sources |
| PraiseProvider & Model Coverage | 0 of 3 positive (under the minimum) | 2 of 4 positive (under the minimum) | 2 of 7 positive | Sources |
| ConcernStability & Code Quality | 11 of 11 raise it | 18 of 18 raise it | 29 of 29 raise it | Sources |
| ConcernPerformance at Scale | 7 of 9 raise it | 7 of 7 raise it | 14 of 16 raise it | Sources |
| ConcernSecurity & Supply Chain | 5 of 6 raise it | 5 of 8 raise it | 10 of 14 raise it | Sources |
| ConcernEnterprise-Gated Features | 2 of 2 raise it (under the minimum) | 4 of 4 raise it (under the minimum) | 6 of 6 raise it | Sources |
| ConcernOperational Burden | 2 of 2 raise it (under the minimum) | 7 of 8 raise it | 9 of 10 raise it | Sources |
| ConsensusOne gateway for the organization | 9 supporting (below threshold) | 5 supporting (below threshold) | 14 supporting (below threshold) | Sources |
| ConsensusStay provider-neutral | 3 supporting (under the minimum) | 8 supporting (below threshold) | 11 supporting (below threshold) | Sources |
| ConsensusCoding tools go through the gateway | 1 supporting (under the minimum) | 18 supporting | 19 supporting (below threshold) | Sources |
| ConsensusThe library alone is enough for some | 1 supporting (under the minimum) | 8 supporting (below threshold) | 9 supporting (below threshold) | Sources |
| ConsensusSome replace it or build their own | 5 supporting (below threshold) | 9 supporting (below threshold) | 14 supporting (below threshold) | Sources |
| DisagreementIs LiteLLM ready to be the production gateway in front of every model call? | Runs it in production 5, Not production-grade 3 | Runs it in production 2, Not production-grade 3 | Runs it in production 7, Not production-grade 6 | Sources |
| DisagreementWhen practitioners weigh LiteLLM against another gateway, which do they keep? | Chose or kept LiteLLM 4, Chose another tool 5 | Chose or kept LiteLLM 5, Chose another tool 11 | Chose or kept LiteLLM 9, Chose another tool 16 | Sources |
| Current readEvidence mix | 59 sources | 90 sources | 137 sources | Sources |
GILD's checks of claims against official documentation, kept apart from the source counts.
