MicrosoftCompanyMicrosoftMicrosoft builds platforms and tools powered by AI, among them Microsoft Copilot and Power Automate. Microsoft says its mission is to help every person and every organization on the planet achieve more.Microsoft CopilotPower AutomateOpen companyAI Model API
Microsoft Foundry
Azure platform, formerly Azure AI Foundry, for building on OpenAI, Anthropic, Meta, DeepSeek and Microsoft models. Foundry Agent Service, evaluations and model router add agents and routing. Usage bills per token or per provisioned unit on Azure.
The 30-Second Take
Pick Microsoft Foundry if you want OpenAI, Anthropic and open models under Entra ID and one Azure agreement.
It sells OpenAI's models directly, plus Claude and open models from Meta, DeepSeek and others. Its data terms win buyers and its model catalog draws praise.
- Best fit
- Teams on Azure that want OpenAI, Anthropic and open models under Entra ID and one Azure agreement.
- Main constraint
- Quota tiers, deployment types and data terms differ by model family, and new subscriptions can start with little or no quota.
- First move
- Pick the deployment type your residency rules allow, then confirm the quota tier for each model you need.
Why
Watch the terms: quota tiers follow usage and agreement status, and deployment types and data terms differ by model family. Requesty's gateway measured 78% request success on Azure in April 2026, so test Agent Service before production.
34 practitioners, 59 sources reviewed, Updated Oct 28, 2026How this is measured
- Operating stack
- AI Application Platform
Models
Which models does Microsoft Foundry serve, and where is each strong?
Each model's page charts its public benchmark scores beside any other model you layer on it.
- Claude Fable 5.1Claude Fable release for demanding reasoning and long-horizon agentic work, with stronger agentic coding, multistep research, and document, spreadsheet, and slide work.
- Claude Haiku 5.5Anthropic's small model for high-volume, latency-sensitive tasks such as classification, extraction, and routing. API model claude-haiku-5-5, 1M-token context.
- Claude Opus 5.5A Claude Opus release for complex agentic, coding, and professional work.
- Claude Sonnet 5.5A Claude Sonnet release for everyday agentic, coding, and knowledge work.
- GPT-6.1 SolOpenAI's newer Sol model, close to GPT-6 Astra on complex work at a lower cost. On OpenAI's own OSWorld 2.1 computer-use run (offline set) it scores 78%, 2.9 points under GPT-6 Astra at about a seventh of the cost per task. API model gpt-6.1-sol, with a 1,050,000-token context on the OpenAI API; Microsoft Foundry caps input at 922,000 tokens.
- GPT-6 AstraOpenAI's highest-capability GPT-6 release for complex end-to-end work.
- GPT-6 LunaOpenAI's most efficient GPT-6 model, for focused, high-volume tasks. API model gpt-6-luna, with a 1,050,000-token context on the OpenAI API; Microsoft Foundry caps input at 922,000 tokens.
- Grok 4.7xAI's Grok release for agentic coding and multistep work. API model grok-4.7, with a 500K-token context window on the xAI API, Microsoft Foundry, and OpenRouter, and 524,288 tokens on Amazon Bedrock and Google Cloud.
Comparisons
How does Microsoft Foundry compare with the alternatives?
Microsoft Foundry vs Amazon Bedrock
- AI Model APIs
Pick the platform where your cloud agreement, identity and data already live. Both sell Claude, OpenAI and open models with batch and reserved capacity; Requesty measured 96% request success on Bedrock and 78% on Azure in April 2026.
Foundry sells OpenAI's GPT models directly; Agent Platform sells Gemini. On one open-model workload, gpt-oss-120b is cheaper on Google Cloud and Llama 4 Maverick is cheaper on Azure.
Operator Insights
What are teams learning from building on Microsoft Foundry and running it in production?
Swarup S. SwaminathanAssociate Professor of Clinical Ophthalmology; the study's senior and corresponding authorBascom Palmer Eye Institute, University of Miami Miller School of Medicine
A University of Miami research team reading medications from glaucoma clinic notes, inside the university's existing Microsoft Azure enterprise environment.
Ophthalmology Science, on PubMed Central, published Read the paper (opens Ophthalmology Science, on PubMed Central in a new tab)
Observation
Bascom Palmer Eye Institute researchers used Microsoft Azure AI Foundry to run five models on 992 glaucoma clinic notes under a Microsoft business associate agreement, and failures ranged from 0% to 0.9%.
- Takeaway
- A hospital can run models from several makers on patient data through one Foundry environment under its Microsoft business associate agreement. Track token use and cost from the first run, which this study did not.
- What happened
- The researchers drew 1,250 glaucoma clinic notes at random, one per patient, from 451,283 eligible notes, excluded 8 supervising-physician attestations to leave 1,242, and used 992 as the held-out test set for every model performance analysis. Claude Opus 4.6, DeepSeek-V3.2, GPT-5.2, Grok 4.1 and Qwen3.6-35B-A3B were accessed through Microsoft Azure AI Foundry inside HIPAA-compliant containers, under a business associate agreement between Microsoft and the University of Miami. On the held-out set, F1 for current topical medications ranged from 0.968 (Grok) to 0.985 (Claude), failures ranged from 0 to 9 (0% to 0.9%) across models and tasks, all from API timeouts or standardization errors, and token usage and costs were not tracked.
- Public engineering write-up (peer-reviewed paper in Ophthalmology Science)
- Measured results on stated settings
- Published Aug 17, 2026
Deployment details
A University of Miami research team reading medications from glaucoma clinic notes, inside the university's existing Microsoft Azure enterprise environment.
- Public engineering write-up (peer-reviewed paper in Ophthalmology Science)
- Measured results on stated settings
- Published Aug 17, 2026
- Team
- Bascom Palmer Eye Institute researchers
- Workload
- Reading current topical and oral medications from glaucoma clinic notes; 992 notes in the held-out test set
- Product version
- Microsoft Azure AI Foundry, in HIPAA-compliant containers within the university's existing Azure enterprise environment
- Model
- Claude Opus 4.6, DeepSeek-V3.2, GPT-5.2, Grok 4.1 and Qwen3.6-35B-A3B
- Measurement
- F1 for current topical medications from 0.968 (Grok) to 0.985 (Claude); failures from 0 to 9 (0% to 0.9%) across models and tasks; token usage and costs not tracked
Chris BakerFurther's About page lists Chris Baker as COO. The post's author says he is a co-founder and signs it With Further Inc.Further Inc.
A 13-person startup in the Microsoft for Startups Founders Hub program, migrating its AI infrastructure.
Microsoft Tech Community, Startups at Microsoft, published Read the post (opens Microsoft Tech Community, Startups at Microsoft in a new tab)
Observation
Further deployed Claude Opus 4.6 and Sonnet 4.6 through Azure AI Foundry assuming its sponsorship credits covered them, and was invoiced $17,493.01 after Foundry failed to warn that these models are excluded from the credits and left their cost estimate blank.
- Takeaway
- Before you deploy Claude or another partner model in Foundry on sponsorship credits, confirm the credits cover it and watch the invoices, because this team found no exclusion warning and no cost estimate for those models.
- What happened
- In February 2026 Further deployed Claude Opus 4.6 and Sonnet 4.6 through Azure AI Foundry as part of a migration of its AI infrastructure, assuming its Azure sponsorship credits covered them. The post says Foundry showed no warning or label that these models are excluded from sponsorship credits, and that the monitoring dashboard's Estimated Cost section was blank for them. Invoices of $1,078.07 for February and $16,414.94 for March followed, $17,493.01 in all; the post says Further could not pay the second, and that Azure Support, Anthropic and the Microsoft for Startups team each referred it elsewhere.
- Public organization write-up (post signed With Further Inc.)
- Posted Apr 13, 2026
Deployment details
A 13-person startup in the Microsoft for Startups Founders Hub program, migrating its AI infrastructure.
- Public organization write-up (post signed With Further Inc.)
- Posted Apr 13, 2026
- Team
- Further Inc., a 13-person startup
- Workload
- A migration of the company's AI infrastructure
- Product version
- Claude models deployed through Azure AI Foundry, expected to run on Azure sponsorship credits
- Model
- Claude Opus 4.6 and Sonnet 4.6
- Measurement
- Invoices of $1,078.07 (February) and $16,414.94 (March), $17,493.01 in all
Buying and deployment
What a team building on it should verify before production traffic grows.
Sourced from Microsoft documentation
Signals
What is the current evidence saying about Microsoft Foundry?
A current read of what teams and builders working in Microsoft Foundry report publicly, from the model catalog to Foundry Agent Service, with each finding linked to its sources.
Practitioner sentiment, all practitioners
−26 Negative net
of 36 sources that take a side, each author counted once.
How this figure is madeTeams and leaders have 8 accounts; GILD prints their sentiment from 20.
Current sentiment
Positive, mixed and negative sources
The score counts them as 9 positive, 7 mixed and 18 negative votes, one per author.
Accounts lean negative. The model catalog and the enterprise data terms draw most of the praise, and contract and data terms are the most common reason given for choosing Foundry. Agent Service failures and missing models or features are the most reported concerns, followed by reliability and a confusing portal.
Run a pilot on your own prompts to measure output quality and cost per task.
Evidence field. All 59 sources, one mark each.
Arrange by
Each mark is one source: positive, mixed, negative, no opinion stated, or documentation (not counted).
| Group | Sources | Positive | Mixed | Negative | No opinion stated and not counted in the score | Documentation, checked and not counted |
|---|---|---|---|---|---|---|
| Positive | 9 | 9 | 0 | 0 | 0 | 0 |
| Mixed | 7 | 0 | 7 | 0 | 0 | 0 |
| Negative | 20 | 0 | 0 | 20 | 0 | 0 |
| Not counted | 9 | 0 | 0 | 0 | 9 | 0 |
| Documentation | 14 | 0 | 0 | 0 | 0 | 14 |
| Quarter | Sources |
|---|---|
| Q1 2025 | 2 |
| Q2 2025 | 0 |
| Q3 2025 | 6 |
| Q4 2025 | 4 |
| Q1 2026 | 3 |
| Q2 2026 | 4 |
| Q3 2026 | 15 |
| Q4 2026 | 11 |
| Documentation, checked | 14 |
By sentiment
14 documentation pages check facts and are not counted.
Hover a mark for a quick look, click it or press Enter to open its source.Tap a mark to open its source, then step to its neighbors with Previous and Next.
What practitioners praise
Top praise of 4. Each mark is a source on it; filled marks praise it.
Most common concerns
Top concern of 6. Each mark is a source on it; filled marks raise it.
All 59 sources behind these findings
- What is Microsoft Foundry?Microsoft (Official), Product documentation, checked Oct 28, 2026
- Authentication and authorization in Microsoft FoundryMicrosoft (Official), Product documentation, checked Oct 28, 2026
- What is Toolbox in Microsoft Foundry?Microsoft (Official), Product documentation, checked Oct 28, 2026
- Microsoft Foundry Models overviewMicrosoft (Official), Product documentation, checked Oct 28, 2026
- Model router for Microsoft Foundry conceptsMicrosoft (Official), Product documentation, checked Oct 28, 2026
- Foundry Models PricingMicrosoft (Official), Product documentation, checked Oct 28, 2026
- Manage traffic with spillover for provisioned deploymentsMicrosoft (Official), Product documentation, checked Oct 28, 2026
- Enable diagnostic logging for Microsoft FoundryMicrosoft (Official), Product documentation, checked Oct 28, 2026
- Azure and other Microsoft cloud services compliance scopeMicrosoft (Official), Product documentation, checked Oct 28, 2026
- System and Organization Controls (SOC) 2 Type 2Microsoft (Official), Product documentation, checked Oct 28, 2026
- ISO/IEC 27001:2013 Information Security Management StandardsMicrosoft (Official), Product documentation, checked Oct 28, 2026
- Data, privacy, and security for Foundry Models sold by Azure in Microsoft FoundryMicrosoft (Official), Product documentation, checked Oct 28, 2026
- Microsoft Foundry Models quotas and limitsMicrosoft (Official), Product documentation, checked Oct 28, 2026
- Migrate from the Foundry (classic) portalMicrosoft (Official), Product documentation, checked Oct 28, 2026
- Anthropic S-1827a, developer (self-described: Discussion of whether enterprises will run their own model hardware), Hacker News comment, published Oct 27, 2026
- VanillaCache on X: "for azure foundry, the platforms really good, but it's a pity I feel like any calls through it are just 2X slower"VanillaCache, developer (self-described: Calls models through Foundry and other hosts), Post on X, published Oct 27, 2026
- Stephen W. Thomas on X: "Microsoft Foundry AI model deployment options confused me!"Stephen W. Thomas, Azure integration architect and Microsoft MVP (self-described: Walkthrough of Foundry deployment types and regional model availability), Post on X, published Oct 27, 2026
- IBM CEO says open weight models are now good and cheap on-premfg137, developer (self-described: Discussion of whether to rent model capacity or buy GPUs), Hacker News comment, published Oct 26, 2026
- Bedrock vs Vertex AI vs Foundry: Your Cloud Commit DecidesRajesh Beri, enterprise AI practitioner and author of THE DAILY BRIEF (self-described: Platform comparison for enterprise buyers choosing where to run Claude, GPT, and Gemini), Newsletter analysis, THE DAILY BRIEF, published Oct 25, 2026
- AiswaryaVenkit1 on X: "At first glance, it can feel like a long list of models, agents, tools, evaluations, guardrails, and governance features."AiswaryaVenkit1, developer (self-described: Thread explaining what Microsoft Foundry includes), Post on X, published Oct 23, 2026
- DeepInfer: High-performance inference framework for DeepSeek modelskasey_junk, developer (self-described: Company whose compliance and procurement teams approve the models developers may use), Hacker News comment, published Oct 20, 2026
- OpenCode 2.0cdnsteve, developer (self-described: Engineering organization of more than 100 developers running Claude Code against its own Foundry deployments), Hacker News comment, published Oct 15, 2026
- [Azure.AI.AgentServer.Responses]
response.completedis streamed before the conversation lock is released, so the next turn gets409 conversation_lockedguramrit-dhillon, developer (self-described: Hosted agent on Foundry Agent Service built with Microsoft Agent Framework on .NET 9), GitHub issue, Azure/azure-sdk-for-net, published Oct 15, 2026 - Why isn't the industry freaking out about DeepSeek 4.1 Flash?browningstreet, developer at a large enterprise (self-described: Large enterprise with a policy on Chinese models), Hacker News comment, published Oct 8, 2026
- Using Opus 5.5 to discover a new eyewitness record of the dodojiggawatts, consultant (self-described: Multi-model OCR workflow that runs the same prompt through several models), Hacker News comment, published Oct 2, 2026
- GPT 5.6 Sol Still Fails on Basic Code Reviewchris_money202, developer (self-described: Compares model behavior across coding tools), Hacker News comment, published Sep 28, 2026
- azure-ai-projects: gpt-6-luna returns 500 on project endpoint, works on resource endpointKonnor (konnor-willison-deepsee), developer (self-described: Python application calling models through a Foundry project in Sweden Central), GitHub issue, Azure/azure-sdk-for-python, published Sep 24, 2026
- .NET: [FoundryAgent][gpt-6-luna]: [Bug]: Invalid parameter: 'text.format' of type 'json_schema' is not supported with model version
gpt-6-luna-2026-09-22lucianteodorescu-milestone, developer (self-described: Foundry agent switched to a newer model), GitHub issue, microsoft/agent-framework, published Sep 24, 2026 - Grok 4.7taspeotis, developer (self-described: Team running non-OpenAI models from its own Foundry resource in the Pi coding harness), Hacker News comment, published Sep 21, 2026
- .NET: [Bug]: Hosted-agent runtime never activates OTLP export to a third-party destination, despite docs describing it as automaticCristopher Coronado, Senior AI Engineer and Microsoft MVP (self-described: .NET hosted agent on Foundry sending telemetry to a third-party destination), GitHub issue, microsoft/agent-framework, published Sep 15, 2026
- DeepSeek v4.1 FlashHanClinto, developer (self-described: Discussion of how to reach models with fewer safety filters), Hacker News comment, published Sep 10, 2026
- DeepSeek v4.1 Flashcdnsteve, developer (self-described: Engineering organization that hosts its coding models in Foundry), Hacker News comment, published Sep 10, 2026
- Navier-Stokes - Tristan Buckmaster [pdf]edot, developer (self-described: Discussion of whether model vendors train on customer data), Hacker News comment, published Sep 9, 2026
- Baptiste GHO on X: "background mode is disabled for every model we've tried in Azure AI Foundry"Baptiste GHO, CTO, Complement Learning (self-described: Course generation feature in production at Complement Learning), Post on X, published Sep 9, 2026
- GPT-6 Astra on OpenRouterjiggawatts, consultant (self-described: Advises customers who choose between Foundry and OpenAI's own API), Hacker News comment, published Sep 5, 2026
- GPT-6 Astrajiocrag, developer (self-described: Discussion of where new OpenAI models are available), Hacker News comment, published Sep 3, 2026
- OpenRouter is joining Stripejohnbarron, developer (self-described: Corporate AI work weighing model routers against cloud platforms), Hacker News comment, published Aug 20, 2026
- Kimi K2.7 Code is generally available in GitHub Copilotpkaye, developer (self-described: Discussion of which models reach Microsoft's platforms), Hacker News comment, published Jul 2, 2026
- Kimi K2.7 Code is generally available in GitHub Copilotrombert, developer (self-described: Discussion of where Chinese open-weight models run in GitHub Copilot), Hacker News comment, published Jul 2, 2026
- Kimi K2.7 Code is generally available in GitHub CopilotmvATM99, developer (self-described: Connects a Foundry model to GitHub Copilot in VS Code), Hacker News comment, published Jul 2, 2026
- I built a WhatsApp clone of myself; a friend agreed on a password with itdandinu, developer (self-described: Personal WhatsApp reply bot with retrieval over the author's past messages), Hacker News comment, published Jun 19, 2026
- Azure AI Foundry Agent published to Microsoft Teams prompts users for Foundry login instead of using Teams SSOJino George, who published a Foundry agent to Microsoft Teams for users in his organization's Microsoft 365 tenant (self-described: A Foundry agent published straight to Microsoft Teams for users in the same Microsoft 365 tenant), Microsoft Q&A thread, published Jun 15, 2026
- Cybersecurity researchers aren't happy about the guardrails on Anthropic's Fableconsumer451, developer building products for clients (self-described: Client products built on Claude models deployed in Azure), Hacker News comment, published Jun 11, 2026
- Decisions that eroded trust in Azure - by a former Azure Core engineermrsmrtss, developer (self-described: Team that has run critical services on Azure for ten years), Hacker News comment, published Apr 3, 2026
- Runtime validation is still fucked in AI coding agentssebringj, developer (self-described: Team trying AI end-to-end testing at the request of the author's manager), Hacker News comment, published Feb 10, 2026
- Create Preview web experience from new AI Foundry is incompleteJan de Vries, engineer, Zure (self-described: Publishing a Foundry agent as a web app from the new portal), GitHub issue, microsoft-foundry/foundry-agent-webapp, published Jan 28, 2026
- [Bug] 400 Error when chatting with an Agent published to Teams/Web using new foundry portalAlexander Danson, Product Manager, Marshall Group (self-described: Agents built in the new Foundry portal and published to Microsoft Teams), GitHub issue, microsoft-foundry/new-foundry-portal, published Jan 9, 2026
- Getting a Gemini API key is an exercise in frustrationhobofan, developer (self-described: Production AI features that need high uptime), Hacker News comment, published Dec 11, 2025
- Getting a Gemini API key is an exercise in frustrationarielcostas, developer (self-described: Project that needed Azure speech voices and Foundry resources), Hacker News comment, published Dec 10, 2025
- Microsoft has a problem: lack of demand for its AI productszubiaur, developer (self-described: Builds on Azure AI services, including AI Search and Document Intelligence), Hacker News comment, published Dec 8, 2025
- Google Antigravitykoakuma-chan, developer (self-described: Enterprise work that must reach models through a cloud platform), Hacker News comment, published Nov 18, 2025
- Cannot deploy on Private Network: Failed to create Aml RP virtual workspace due to System.Exception: Failed async operationDaryl Graves, developer (self-described: Deploying the standard agent setup on a private network from Microsoft's template), GitHub issue, published Sep 23, 2025
- The LLM Lobotomy?Julian Goldsmith (juliangoldsmith), developer (self-described: Ongoing project with low request volume on Foundry), Hacker News comment, published Sep 20, 2025
- azure-ai-agents connected agent with MCP will raise for approval and eventually crashIlya Rakhlin (irakhlin), developer (self-described: Multi-agent setup with a coordinator agent and a connected agent, each with an MCP tool), GitHub issue, Azure/azure-sdk-for-python, published Aug 17, 2025
- azure-ai-agents: MCP usage errors with Azure AI Foundryreesmonty (jacobreesmontgomery), developer (self-described: Foundry agents calling remote MCP servers such as GitMCP and Tavily), GitHub issue, Azure/azure-sdk-for-python, published Aug 8, 2025
- Azure AI Foundry Agent Service Deep Research Tool ErrorSashank Neelamraju (sashanknjs), developer (self-described: Trying the Deep Research tool in Foundry Agent Service), GitHub issue, Azure/azure-sdk-for-python, published Jul 8, 2025
- Tool call of AzureAISearchTool doesn't respect tool configuration such as
filterandtop_kin packageazure-ai-agentsVitor Arrais, engineer, N26 (self-described: Retrieval agent over an Azure AI Search index), GitHub issue, Azure/azure-sdk-for-python, published Jul 4, 2025 - 400 reasons to not use Microsoft AzureOras, developer (self-described: A few months building on Azure AI Foundry and AI Search), Hacker News comment, published Mar 1, 2025
- DeepSeek R1 Is Now Available on Azure AI Foundry and GitHubvemonet, developer (self-described: Testing DeepSeek R1 and Mistral Large through serverless APIs on a paid subscription), Hacker News comment, published Jan 30, 2025
Jan 2025 to Oct 2026.
45 sources from practitioners, by month and sentiment.
Where It Fits
Where does Microsoft Foundry belong in the stack?
Microsoft Foundry is Core across Model and Operate and supports Build, Data, and Evaluate.
- Business area
- Product & Engineering
- Function
- Software Development
- Operating stack
- AI Application Platform
- Primary role
- Managed Model Access & Cloud AI Platform
Model
Model APIs: Core
Model APIs, core. Foundry Models lists more than 10,000 models from Microsoft, OpenAI, Anthropic, Meta and others, with Azure OpenAI models sold directly by Azure. Usage bills per token or per provisioned throughput unit on Azure.
Model gateway: Supports
Model gateway, supports. Model router is a deployable model that sends each prompt to the most suitable underlying model, and spillover moves excess provisioned traffic to a standard deployment. Use LiteLLM or OpenRouter for models outside Foundry.
Build
Orchestration: Supports
Orchestration, supports. Foundry Agent Service runs declarative prompt agents and hosted agents that run the team's own code, including Microsoft Agent Framework, LangGraph and Semantic Kernel. The framework itself stays the team's choice.
Tool use: Supports
Tool use, supports. Agents call MCP servers, OpenAPI tools, Azure Logic Apps connectors and knowledge tools for Azure AI Search, SharePoint and Microsoft Fabric. Tools outside Azure still need their own hosting.
Data
Retrieval: Supports
Retrieval, supports. Foundry IQ, Azure AI Search, File Search and Grounding with Bing Search cover managed retrieval for agents. Keep a separate vector store for custom retrieval.
Evaluate
Evaluation: Supports
Evaluation, supports. Built-in evaluators score agents for coherence, relevance, groundedness and safety. Use LangSmith or Braintrust to compare evaluations across clouds and model vendors.
Operate
Observability: Supports
Observability, supports. Foundry traces agent interactions with OpenTelemetry and sends resource logs to Azure Monitor. Use an observability tool for models and agent steps outside Azure.
Cloud deployment: Core
Cloud deployment, core. Inference runs inside the Azure subscription with Entra ID, role-based access control and network isolation. Global, Data Zone and regional deployment types, quota tiers and provisioned throughput set residency, price and capacity.
Guardrails: Supports
Guardrails, supports. Foundry Guardrails, previously content filters, run synchronously as the service processes prompts and generates content. Prompt-injection testing and red-teaming stay with a dedicated security product.
Market mapThe layers Microsoft Foundry covers, and who else owns each one.
Model APIs
Cloud deployment & capacity
- Amazon Bedrock
- Fireworks AI
- Gemini Enterprise Agent Platform
- Together AI
- Microsoft Foundry
- Groq
Model gateways & routing
- LiteLLM
- OpenRouter
- Portkey
- Vercel AI Gateway
Orchestration & agent frameworks
Tool use & integrations
4 more layers on the full map. Core, Supports and Touches describe coverage. They say nothing about quality.

