Microsoft used its Ignite 2026 conference, which began on November 17, to announce a batch of changes to Microsoft Foundry, the platform it formerly called Azure AI Foundry. Some items are generally available, which means any customer can use them, and others are only in preview. For a team deciding what to build on, that difference matters more than the headline list. This article sorts the Foundry announcements by status, using Microsoft’s own Book of News and an independent write-up, and notes what neither source says. Status is as published on November 17 and 18 and may have changed since.
Generally available now
- Multi-turn evaluation. The Book of News says it “assesses entire conversations against quality standards”, so agent chats can be scored as a whole rather than one reply at a time.
- Production insights. Foundry “automatically analyzes production signals” with a new insights capability, according to the same document.
- Agent optimizer. Per the Book of News, it proposes candidate changes and tests them against a team’s own evaluations. SiliconANGLE’s write-up does not mention the optimizer or multi-turn evaluation, so these two rest on Microsoft’s document alone.
- Microsoft-Decision-1.1. Generally available in Foundry, OpenRouter and Vercel, following the first Decision model in October. SiliconANGLE says it is based on Google’s Gemma 4 26B A4 with added multimodal support and is offered at a “competitive price”; neither source gives a price.
- MAI-Code-2 and MAI-Code-2-Flash. Generally available on GitHub. In Foundry, Code-2-Flash is in private preview, not general availability.
Preview only
- Human annotation, which lets experts rate and comment on agent traces.
- Managed Training and Programmed Training. The first adapts supported models with organisational data; the second lets AI engineers write their own training loops.
- MAI-Image-2.7 and MAI-Image-2.7-Flash.
- MAI-Cyber-2-Flash, limited to eligible organisations inside Codename MDASH.
- Video models from fal (Kling, LTX and MiniMax) and Baseten serverless deployments, which the Book of News says are offered “globally or in the US Data Zone”.
- Foundry IQ items: knowledge graphs, Microsoft-Retriever-1 and a Foundry IQ resource in every new project. The SharePoint indexed knowledge source is generally available.
- Foundry on AKS: Agent Grid, the Agent Sandbox extension for AKS and Flex nodes. The Book of News says all capabilities in its Azure section are in preview except Azure Cobalt 200.
What Microsoft did not say
- Prices. The sources we read list no per-token or per-hour prices for the new models or features, apart from the “competitive price” wording for Decision-1.1.
- Regions. Only Baseten’s deployment options mention geography.
- Benchmarks. Neither source publishes test results for the new models, so any performance claim should wait for independent testing.
- Model router. The Book of News mentions an expanded pool of the top 50 models, but gives no availability status.
What to check before building on it
- Read the status label, not the headline. “Preview” features can change or be withdrawn and often come without a service-level agreement.
- Check the Foundry documentation for your region and your data-zone requirements.
- Ask for pricing before piloting anything that depends on a new model.
- Test evaluation features on a non-critical agent first, since they judge the quality of your own workflows.
For related platform news, see our coverage of Microsoft 365 Copilot usage-based billing.
Sources
- Microsoft: Ignite 2026 Book of News (Microsoft Foundry section)
- SiliconANGLE: Microsoft rolls out new first-party models and updates to agentic AI continuous improvement (Nov 17)
Image: Jonathan Schilling / CC BY-SA 4.0 via Wikimedia Commons.
Like our coverage? Add stepphase.com as a preferred source on Google to see more of our stories in Top Stories.
