Skip to main content

Gemini API vs Amazon Bedrock: One Model Family or a Model Marketplace?

Slack and Wealthsimple run both, routing by job. Bedrock does not sell Gemini. Team accounts, a monthly bill at one workload, scores, batch, capacity, data terms, and model lifecycle compared.

Reading time
7 min read
Sources
33 cited
In this piece

GILD verdict

Choose Amazon Bedrock on AWS for several model families under IAM and AWS billing, though it does not sell Gemini.

Teams that run both route by job: Slack runs Slack AI on Amazon Bedrock and added Google Cloud for Gemini in early 2026, and Wealthsimple added Gemini for long inputs before it added Bedrock. Ziva.sh's throttled account shows the quota has to be secured before launch. Choose the Gemini API when your evaluations favor Gemini on long or multimodal input. At GILD's 60,000-task workload, Gemini 4 Argon costs US$1,602 a month until its introductory price ends on November 30, 2026 and US$3,204 after, against US$1,632 for Claude Sonnet 5.5 on Bedrock with caching. Budget for Flash's introductory price ending on December 31, 2026 and for Gemini's yearly model retirements.

Teams that buy both route by job: Slack runs Slack AI on Amazon Bedrock and in early 2026 added Google Cloud's Vertex AI for Gemini and other models, and Wealthsimple added Gemini to its gateway for 1 million token inputs before it added Bedrock. The Gemini API sells Google's Gemini models through an AI Studio key or Google Cloud. Bedrock is AWS's catalog of more than 100 models from vendors such as Amazon, Anthropic, OpenAI, and xAI, with Google present only through its Gemma open models. At GILD's workload of 60,000 cached tasks a month, Gemini 4 Argon costs US$1,602 until its introductory price ends on November 30, 2026, and Claude Sonnet 5.5 on Bedrock US$1,632. One model-level price example: Claude Haiku 5.5 on Bedrock outscores Gemini 3.8 Flash on the Artificial Analysis Intelligence Index v5.0 (Oct 27, 2026), 36 to 31, at about a nineteenth of Flash's cost per task, US$0.41 against US$7.77.

What teams report

Failures and incidents 1

  1. Ziva.shOn Amazon Bedrock

    Ziva.sh, a small SaaS running Claude models on Bedrock, found its account capped about 1,000 times below the default quota, with 14% of Haiku and 16% of Opus requests throttled over 14 days, and AWS denied its increase request (May 16, 2026).

    “TL;DR: AWS is throttling our Bedrock usage at 1,000x below the minimum quota and won’t increase it.”
    Ziva.sh team, Company blog, Ziva.shZiva.sh blog, , write-up

Run both 3

  1. Slack (Salesforce)

    Slack moved Slack AI's model serving to Amazon Bedrock from mid-2024 and, by early 2026, added Google Cloud's Vertex AI for Gemini and other models beyond failover, reporting about 10% better quality on complex reasoning tasks (Engineering at Slack, May 28, 2026).

    “By early-2026 we officially expanded our footprint to include Google Cloud Platform (GCP) Vertex AI, not just as a failover for redundancy, but as a strategic engine to accelerate product innovation through access to a broader catalog of state-of-the-art models.”
    Shaurya Kethireddy, Lead Member of Technical Staff, Slack (Salesforce)Engineering at Slack, , write-up

  2. Wealthsimple

    Wealthsimple's internal LLM gateway added Gemini beside OpenAI and Cohere for its 1 million token context, then added Amazon Bedrock after multimodal support, and in September 2024 moved its API to OpenAI's format (InfoQ, Feb 6, 2025).

    “After supporting multi-modal inputs, the next thing we added to our platform was Amazon Bedrock. Bedrock is AWS's managed service for interacting with foundational large language models.It also provides the ability to deploy and fine-tune these models at scale.”
    Mandy Gu, Author of InfoQ's article on Wealthsimple's generative AI tools, WealthsimpleInfoQ, , write-up

  3. DoorDashOn Gemini API

    DoorDash runs Gemini beside OpenAI and Anthropic models behind one gateway, with Amazon Bedrock as the Claude API's fallback. Its platform lead names cost at scale and Gemini's one-year model lifetime, which forces hundreds of use cases to migrate, as the cracks that followed (InfoQ talk, Oct 3, 2026).

    “Similarly, on the model side, we went vendor first. Let's use OpenAI, Anthropic, Gemini models. Folks started getting productive. Folks learned how to do prompt engineering. They were able to scale up. At DoorDash scale, at some point, cost becomes an issue. There were a few other cracks that started to appear. Like the one is the deprecation treadmill. How many of you know that Gemini has a one-year lifetime for models? In one year, the model gets shut down, and this applies to every new model. At the end of that one year, you have to switch to a newer model, whether or not it meets your business needs. I could probably accept that if I was a single product team and I was working on my product, but I as a platform team, I have to think about hundreds of use cases and how I'm going to drive hundreds of teams to migrate to the latest model without necessarily giving them a business argument of the benefits.”
    Swaroop Chitlur, Lead, Generative AI Platform, DoorDashInfoQ, , recording

Adopted 1

  1. TinesOn Amazon Bedrock

    Tines built its first AI features on Amazon Bedrock to keep customer data in each tenant's region and out of model training, and keeps a Bedrock translation layer beside OpenAI-compatible providers (Tines blog, Mar 10, 2026).

    “The first AI features we launched were built on AWS Bedrock in order to always maintain customer data within the same geographic regions as their tenant, as well as out of any LLM training pipelines.”
    Yannick Gloster, Software Engineer, TinesTines blog, , write-up

Choosing between Gemini API and Amazon Bedrock

Choose Gemini API if

  • Your evaluations favor Gemini Flash or Flash-Lite on price, or on image, audio, video, and PDF input.
  • You want Google's flagship reasoning at its introductory price, which ends on November 30, 2026: Gemini 4 Argon scores 50 at US$3.84 per task on the Artificial Analysis Intelligence Index v5.0 (Oct 27, 2026), below GPT-6 Astra's US$6.24 per task and its score of 57.
  • Your workload needs long or multimodal inputs that your current provider handles worse, the reason Wealthsimple added Gemini's 1 million token context models to its gateway (Feb 6, 2025).
  • You want to prototype on a free tier before attaching billing, and move to paid-tier terms for customer data.
  • Your organization runs on Google Cloud, or can call an external API from its AWS workloads.

Choose Amazon Bedrock if

  • Your organization runs on AWS and wants model access under IAM, VPC endpoints, and AWS billing.
  • You want several model families, such as Claude, Amazon Nova, and OpenAI models, behind one AWS account.
  • You need reserved capacity or geographic pinning through Bedrock's Reserved tier and cross-Region inference profiles. Secure the quota early: Ziva.sh found its new account throttled 1,000 times below the default quota (May 16, 2026).
  • You need customer data kept in the tenant's region and out of model training, the reason Tines built its first AI features on Bedrock (Mar 10, 2026).

Side by side

DimensionGemini APIAmazon Bedrock
Monthly bill at one workloadGILD's calculation from published rates, fetched Oct 28, 2026: 2,000 tasks a day for 30 days, each with 2,000 new input tokens, 1,000 cache-write tokens, 7,000 cache-read tokens and 2,000 output tokens, at list prices. Gemini's implicit caching bills the 1,000 cache-write tokens as ordinary input. The uncached Bedrock line sends 10,000 input and 2,000 output tokens per task. Taxes and negotiated discounts are excluded.US$1,602 a month on Gemini 4 Argon through November 30, 2026; US$3,204 at its standard US$4 input and US$20 output from December 113US$1,632 a month on Claude Sonnet 5.5 with caching; US$2,400 without caching25
What teams reportDoorDash names Gemini's one-year model lifetime as a migration cost across hundreds of use cases (Oct 3, 2026); Wealthsimple added Gemini for 1 million token inputs (Feb 6, 2025); Slack added Google Cloud for Gemini beside Bedrock in early 2026123Slack serves Slack AI to millions of users on Bedrock (May 28, 2026); Tines chose it to keep data in each tenant's region (Mar 10, 2026); Ziva.sh's account was throttled 1,000 times below the default quota and its increase denied (May 16, 2026)145
ModelsGoogle's models only: Gemini 4 Argon, 3.8 and 3.6 Flash, the Flash-Lite models, the pre-release 3.1 Pro, and Gemini 2.5 for existing users, plus Live, image, speech, music (Lyria), and embedding modelsMore than 100 models from providers including Amazon, Anthropic, DeepSeek, OpenAI, and xAI; Google is represented by Gemma open models
Workhorse priceIntroductory paid-tier price, then the standard price, in US dollars.Gemini 3.8 Flash US$0.75 input and US$3.75 output per million tokens through December 31, 2026, then US$1.50 and US$7.50Set per model and per Region on the Bedrock pricing page; cross-Region inference adds no routing charge
Independent scoresArtificial Analysis Intelligence Index v5.0 (Oct 27, 2026) at each model's highest-scoring reasoning effort, high for the Gemini models and max for the others, with its average API cost per index task at the model makers' list prices, checked Oct 28, 2026.Gemini 4 Argon 50 at US$3.84 per task; Gemini 3.8 Flash 31 at US$7.77689Claude Opus 5.5 58 at US$10.04 per task; GPT-6 Astra 57 at US$6.24; Claude Haiku 5.5 36 at US$0.416101112
Discounted and premium processingBatch and Flex at 50% off standard; Priority at 1.8x the standard rate on 3.8 FlashBatch at 50% off on-demand for select models; Flex at 50% off and Priority at a 75% premium over Standard
Batch limitsBatch runs on the generateContent API with a 24-hour target turnaroundBatch inference does not support tool calling or structured output
Reserved capacityProvisioned Throughput on the Agent Platform, a fixed-cost, fixed-term subscriptionReserved tier through the AWS account team, and Provisioned Throughput billed hourly by model unit, with no commitment or 1- or 6-month terms
Data usePaid use is not used to improve Google's products, free-tier use is, and prompts are kept 55 days for abuse monitoringInputs and outputs are not used to train any model and are not stored by default. Flagged OpenAI traffic and all Claude Fable traffic are kept up to 30 days for abuse detection
Model lifecycleEarliest shutdown dates on the deprecations page, with some retired Flash versions routed to successors; Gemini 2.5 models retired on the Agent Platform on October 20, 2026A Legacy period of 6 months or 45 days for models launched from September 7, 2026, on dates that can differ from the provider's
Where it runsThe Gemini Developer API, and Google Cloud's Agent Platform with global, multi-region, and regional endpointsAWS Regions, with Geographic and Global cross-Region inference profiles

What practitioners report, side by side

Each product's figures come from its own practitioners, for the group you pick.

Gemini APIGoogle0 MixedHow this figure is made
Amazon BedrockAmazon Web Services+9 MixedHow this figure is made
Whose sources to count
Compare Gemini API with
  • Amazon Bedrock

Overall signal

Gemini API reads 0 and Amazon Bedrock +9, each from its own practitioners. Gemini API: 24 positive, 25 negative of 74. Amazon Bedrock: 25 positive, 20 negative of 57.

The score counts the authors from teams and leaders that take a side, one vote each with their latest view, and is how far positive votes lead negative ones, per 100 votes, so Gemini API's 0 rests on votes split 23 positive, 24 mixed and 23 negative.

Each of Amazon Bedrock's authors wrote one source here, so each source is one vote. A lead of a fifth of the votes reads positive and a lead of half strong positive, with the same lines below zero. The dots show each source, positive, mixed, negative. Run a pilot on your own prompts to measure output quality and cost per task.

Footprint on the AI Application Platform stack

They differ on five of 9 layers. Gemini API owns one that Amazon Bedrock does not, and Amazon Bedrock owns one that Gemini API does not. Coverage is each product's placement on this stack, as its product page shows it. Core is a job the product owns; Supports and Touches are jobs it helps with. Coverage says nothing about how well a product does them.

Layers where Gemini API and Amazon Bedrock differ
LayerGemini APIAmazon Bedrock
Model
Model gatewayTouchesSupports
Build
Tool useCoreSupports
Evaluate
EvaluationTouchesSupports
Operate
ObservabilityTouchesSupports
Cloud deploymentSupportsCore
The 4 layers where they match
Layers where Gemini API and Amazon Bedrock match
LayerGemini APIAmazon Bedrock
Model
Model APIsCoreCore
Build
OrchestrationSupportsSupports
Data
RetrievalSupportsSupports
Operate
GuardrailsSupportsSupports

What practitioners praise

Among teams and leaders, Gemini API alone has five praise themes that reach 5 sources from 3 authors. Multimodal & Long Context and Model Quality lead, at 13 of 13 and 11 of 12. Both carry Cost Efficiency. It is praised in 16 of the 17 Gemini API sources that discuss it, against 5 of 5 for Amazon Bedrock. Amazon Bedrock alone has three praise themes that reach 5 sources from 3 authors. Model Choice and AWS Integration lead, at 9 of 11 and 7 of 7.

What practitioners praise, Teams and leaders
Praised forGemini APIAmazon Bedrock
Cost Efficiency16 of 175 of 5
Multimodal & Long Context13 of 13Not a theme in its sources
Model Quality11 of 124 of 4 so far
Speed6 of 92 of 2 so far
Tool Use & Structured Output4 of 91 of 2 so far
Developer Experience3 of 60 of 2 so far
Model ChoiceNot a theme in its sources9 of 11
AWS IntegrationNot a theme in its sources7 of 7
Managed Agents, RAG & GuardrailsNot a theme in its sources3 of 6

Each count is the sources that praise a theme, or raise it as a concern, out of those that discuss it. A row pairs the two products when both carry a theme under the same or an equivalent name.

A count ending in "so far" has under 5 sources from 3 authors among teams and leaders. "Not a theme in its sources" means GILD found no theme of that name in that product's sources.

What practitioners raise as concerns

Among teams and leaders, Gemini API alone has six concern themes that reach 5 sources from 3 authors. Model Changes and Latency lead, at 14 of 14 and 11 of 11. Amazon Bedrock alone has five concern themes that reach 5 sources from 3 authors. Quotas, Capacity & Latency and Setup & Model Access lead, at 12 of 15 and 7 of 8.

What practitioners raise as concerns, Teams and leaders
Raised as a concernGemini APIAmazon Bedrock
Model Changes14 of 14Not a theme in its sources
Latency11 of 114 of 4 so far
Billing & Spend Control9 of 9Not a theme in its sources
Reliability & Capacity9 of 9Not a theme in its sources
Rate Limits & Tiers8 of 8Not a theme in its sources
Two Platforms & Setup5 of 5Not a theme in its sources
Quotas, Capacity & LatencyNot a theme in its sources12 of 15
Setup & Model AccessNot a theme in its sources7 of 8
Data Terms & RetentionNot a theme in its sources7 of 10
Cost SurprisesNot a theme in its sources4 of 6
Regional AvailabilityNot a theme in its sources2 of 5
Gemini API SignalsAmazon Bedrock Signals

33 sources

  1. Engineering at Slack: Slack AI, the path to multi-cloud (May 28, 2026)Public sourceslack.engineering
  2. InfoQ: Mandy Gu on Wealthsimple's LLM gateway and productivity tools (Feb 6, 2025)Public sourceinfoq.com
  3. InfoQ: Building a GenAI Platform at DoorDash, talk by Swaroop Chitlur and Siddharth Kodwani (Oct 3, 2026)Public sourceinfoq.com
  4. Ziva.sh: 16 days, 4 channels, no AWS Bedrock quota fix (May 16, 2026)Public sourceziva.sh
  5. Tines: Futureproofing Tines, designing AI infrastructure for scale (Mar 10, 2026)Public sourcetines.com
  6. Artificial Analysis: LLM Leaderboard (Oct 28, 2026)Public sourceartificialanalysis.ai
  7. Artificial Analysis: Changelog (Oct 28, 2026)Public sourceartificialanalysis.ai
  8. Artificial Analysis: Gemini 4 Argon analysis (Oct 28, 2026)Public sourceartificialanalysis.ai
  9. Artificial Analysis: Gemini 3.8 Flash analysis (Oct 28, 2026)Public sourceartificialanalysis.ai
  10. Artificial Analysis: Claude Opus 5.5 analysis (Oct 28, 2026)Public sourceartificialanalysis.ai
  11. Artificial Analysis: GPT-6 Astra analysis (Oct 28, 2026)Public sourceartificialanalysis.ai
  12. Artificial Analysis: Claude Haiku 5.5 analysis (Oct 28, 2026)Public sourceartificialanalysis.ai
  13. Gemini API docs: Gemini Developer API pricingOfficial documentationai.google.dev
  14. Gemini API docs: ModelsOfficial documentationai.google.dev
  15. Gemini API docs: Batch APIOfficial documentationai.google.dev
  16. Gemini API docs: Flex inferenceOfficial documentationai.google.dev
  17. Gemini API docs: Priority inferenceOfficial documentationai.google.dev
  18. Gemini API Additional Terms of ServiceOfficial documentationai.google.dev
  19. Gemini API docs: Abuse monitoringOfficial documentationai.google.dev
  20. Gemini API docs: Gemini deprecationsOfficial documentationai.google.dev
  21. Agent Platform docs: Model versions and lifecycleOfficial documentationdocs.cloud.google.com
  22. Agent Platform docs: Provisioned Throughput overviewOfficial documentationdocs.cloud.google.com
  23. Agent Platform docs: Deployments and endpointsOfficial documentationdocs.cloud.google.com
  24. Amazon Bedrock User Guide: OverviewOfficial documentationdocs.aws.amazon.com
  25. Amazon Bedrock pricingOfficial documentationaws.amazon.com
  26. Amazon Bedrock User Guide: Batch inferenceOfficial documentationdocs.aws.amazon.com
  27. Amazon Bedrock User Guide: Service tiersOfficial documentationdocs.aws.amazon.com
  28. Amazon Bedrock User Guide: Provisioned ThroughputOfficial documentationdocs.aws.amazon.com
  29. Amazon Bedrock User Guide: Cross-Region inferenceOfficial documentationdocs.aws.amazon.com
  30. Amazon Bedrock FAQsOfficial documentationaws.amazon.com
  31. Amazon Bedrock User Guide: Abuse detectionOfficial documentationdocs.aws.amazon.com
  32. Amazon Bedrock User Guide: Model lifecycleOfficial documentationdocs.aws.amazon.com
  33. Amazon Bedrock User Guide: Data protectionOfficial documentationdocs.aws.amazon.com
Read the research