model-releases

# Microsoft Foundry Model Releases

> Release capsules for Microsoft Foundry model announcements — each pairs a model release with a runnable notebook. A capsule is a folder holding a README and a Jupyter notebook you can run against your own Foundry project.

Announcements are tracked in CHANGELOG.md, newest first, grouped by month. Capsules are indexed in CAPSULE-TOC.md, grouped by provider. catalog.json holds this same index in structured form.

## Capsules

- [MAI-Transcribe-2](https://github.com/microsoft-foundry/model-releases/tree/main/models/microsoft-ai/mai-transcribe-2): Transcribe noisy multilingual audio with diarization, timestamps, and domain biasing. Publisher: microsoft-ai. Capabilities: audio-speech. Released 2026-09-03.
- [Model Router — August 2026](https://github.com/microsoft-foundry/model-releases/tree/main/models/model-router/aug-2026): Catch up on routing modes, model subsets, deployment types, failover, and agentic routing as optimization levers. Publisher: model-router. Capabilities: model-router, chat-completion, function-calling. Released 2026-08-19.
- [MAI-Image-2.5-Pro](https://github.com/microsoft-foundry/model-releases/tree/main/models/microsoft-ai/mai-image-2.5-pro): Render portraits, in-image text, and complex scenes. Publisher: microsoft-ai. Capabilities: image-generation. Released 2026-07-23.
- [MAI-Voice-2](https://github.com/microsoft-foundry/model-releases/tree/main/models/microsoft-ai/mai-voice-2): Direct expressive, multilingual, and long-form speech synthesis. Publisher: microsoft-ai. Capabilities: audio-speech. Released 2026-06-02.
- [MAI-Transcribe-1.5](https://github.com/microsoft-foundry/model-releases/tree/main/models/microsoft-ai/mai-transcribe-1.5): Transcribe multilingual audio with keyword biasing. Publisher: microsoft-ai. Capabilities: audio-speech. Released 2026-06-02.
- [MAI-Image-2.5-Flash](https://github.com/microsoft-foundry/model-releases/tree/main/models/microsoft-ai/mai-image-2.5-flash): Generate images faster and at lower cost. Publisher: microsoft-ai. Capabilities: image-generation. Released 2026-06-02.
- [MAI-Image-2.5](https://github.com/microsoft-foundry/model-releases/tree/main/models/microsoft-ai/mai-image-2.5): Generate and edit images from text prompts. Publisher: microsoft-ai. Capabilities: image-generation. Released 2026-06-02.

## Multi-model scenarios

- [MAI-Image-2.5 family walkthrough](https://github.com/microsoft-foundry/model-releases/tree/main/models/microsoft-ai/multi-model-scenarios/mai-image-family-walkthrough): Compare MAI-Image-2.5, Flash, and Pro on identical prompts. Compares: mai-image-2.5-flash, mai-image-2.5, mai-image-2.5-pro.

## Capability primers

- [audio-speech](https://github.com/microsoft-foundry/model-releases/blob/main/docs/primers/audio-speech.md): STT, TTS, or realtime voice.
- [chat-completion](https://github.com/microsoft-foundry/model-releases/blob/main/docs/primers/chat-completion.md): General instruction-following and dialogue.
- [embeddings](https://github.com/microsoft-foundry/model-releases/blob/main/docs/primers/embeddings.md): Vector representations for retrieval / similarity.
- [fine-tuning](https://github.com/microsoft-foundry/model-releases/blob/main/docs/primers/fine-tuning.md): Supports customization / distillation.
- [function-calling](https://github.com/microsoft-foundry/model-releases/blob/main/docs/primers/function-calling.md): Structured tool invocation.
- [image-generation](https://github.com/microsoft-foundry/model-releases/blob/main/docs/primers/image-generation.md): Text → image output.
- [long-context](https://github.com/microsoft-foundry/model-releases/blob/main/docs/primers/long-context.md): 200k+ token context window.
- [model-router](https://github.com/microsoft-foundry/model-releases/blob/main/docs/primers/model-router.md): One endpoint that routes requests across models.
- [multimodal](https://github.com/microsoft-foundry/model-releases/blob/main/docs/primers/multimodal-models.md): Accepts image (and/or audio) input alongside text.
- [reasoning](https://github.com/microsoft-foundry/model-releases/blob/main/docs/primers/reasoning-models.md): Extended-thinking / chain-of-thought optimized models.

## Optional

- [CHANGELOG](https://github.com/microsoft-foundry/model-releases/blob/main/CHANGELOG.md): every Foundry model release announcement
- [CAPSULE-TOC](https://github.com/microsoft-foundry/model-releases/blob/main/CAPSULE-TOC.md): capsule index grouped by provider
- [Glossary](https://github.com/microsoft-foundry/model-releases/blob/main/docs/GLOSSARY.md): terminology used across capsules
- [Quickstart](https://github.com/microsoft-foundry/model-releases/tree/main/models/quickstart): one-time Foundry project and .env setup