In April 2026, Microsoft launched the MAI (Microsoft AI) Labs family of models, designed to offer high-performance alternatives to third-party APIs at significantly lower compute costs.
| Model | Capability | Key Advantage |
|---|---|---|
| MAI-Transcribe-1 | Speech Recognition | High accuracy across 25 languages at a fraction of the GPU cost of Whisper. |
| MAI-Voice-1 | Speech Generation | High-fidelity custom voice creation from very short audio clips. |
| MAI-Image-2 | Text-to-Image | Extreme visual fidelity with lightning-fast generation speeds. |
| harrier-oss-v1 | Text Embeddings | Multilingual open-source embedding family optimized for semantic search. |