Find the right model for your next agentic task.
Changelog • Capsules • Quickstart • Reference
The Microsoft Foundry catalog has thousands of models from Anthropic, Microsoft, OpenAI, xAI, Hugging Face, Meta, Mistral, Cohere, and NVIDIA - and new releases land almost daily. Keeping up is one thing. Knowing what a release actually does differently, and whether it's the right pick for your agent, is another.
That's what this repo is for:
- Catching up? The CHANGELOG has every announcement in one place.
- Want to try one? Browse models/ for a release capsule - a runnable notebook that shows you what it does.
- Can't find the one you need? Open an issue and we'll backfill it.
Every new model release announcement gets a row here, with a link back to the original post plus the publisher and capabilities at a glance. The three most recent:
| Model | Release date | Capabilities |
|---|---|---|
| aug-2026 | 2026-08-19 | Model Router · Chat Completion · Function Calling |
| MAI-Thinking-1 | 2026-08-12 | Reasoning · Chat Completion · Function Calling · Long Context |
| MAI-Code-1.1-Flash | 2026-08-11 | Chat Completion |
See the full CHANGELOG for everything else.
An announcement tells you a model exists. It doesn't tell you what it's like to use. So each capsule pairs the release with a notebook you can run - practical use cases, code-first - to show you:
- What tasks the publisher's models are good at
- What new features this release brings
- What tradeoffs it makes on cost, quality, and latency
Start with the Quickstart, then pick one:
| Capsule | Last updated | Description |
|---|---|---|
| Model Router — August 2026 | 2026-08-21 | Catch up on routing modes, model subsets, deployment types, failover, and agentic routing as optimization levers |
| MAI-Voice-2 | 2026-08-19 | Direct expressive, multilingual, and long-form speech synthesis |
| MAI-Image-2.5 | 2026-08-11 | Generate and edit images from text prompts |
See the full CAPSULE-TOC for everything else.
Want to run a capsule notebook? Four steps, and the first two are one-time setup:
- Open in GitHub Codespaces - you get a ready-to-run environment, no local install.
- Do the quickstart once, to set up a Foundry project and your
.env. - Pick a capsule -
models/is organized by provider, then release. Ex:models/microsoft-ai/mai-image-2.5/. - Open its notebook in VS Code, select the kernel, and run.
Each notebook is yours to break - change the prompts, swap the inputs, and try it against your own scenario.
Want the background - how this repo is organized, what the publishers and capability tags mean, or how to contribute a capsule of your own? It's all in the reference guide.
| Topic | What you'll find |
|---|---|
| Repository structure | How docs/, models/, and the capsules fit together |
| Publishers | Every publisher we track, with links |
| Model capabilities | What each capability tag means, plus a primer for each |
| Contributing | Add a capsule, publisher, capability, or glossary term |
| Using this repo from an agent | catalog.json and llms.txt — the whole catalog in one fetch |
New to the terminology? Start with the glossary.