OpenRouter
@OpenRouter
Three new @MicrosoftAI models now live on OpenRouter! Launching together: MAI-Image-2.5, MAI-Transcribe-1.5, and MAI-Voice-2. More on each below 🧵 https://t.co/KD5JlX6DT6
3retweets40likes
View on XOpenRouter adds Microsoft MAI models for high speed multimodal generation
OpenRouter has added three specialized models from the newly launched Microsoft AI (MAI) family. This release includes
MAI-Image-2.5 for generation and editing, MAI-Transcribe-1.5 for speech-to-text, and MAI-Voice-2 for expressive text-to-speech. These models are built without distillation—training a smaller model to mimic a larger one—to ensure original performance.- Voice Language Support
- 15 languages
- Transcription Performance
- 1 hour of audio in under 15 seconds
- Voice Control Options
- Excited, embarrassed, and whispered
- Alternative Hosting
- Fireworks and Baseten
This integration follows the recent MAI-Image-2.5 debut on the Arena image leaderboard. By hosting these on OpenRouter, Microsoft makes its frontier-grade media capabilities accessible outside of Azure. This move follows how OpenRouter previously added the xAI creative stack to provide a single endpoint for multimodal workflows.
Developers can now use the OpenRouter API for production-grade audio and visual tasks. MAI-Transcribe-1.5 transcribes an hour of audio in under 15 seconds, while MAI-Voice-2 offers emotional controls like whispering. These models are available now alongside hundreds of others via a unified API.
Still wondering? A few quick answers below.
OpenRouter has added three models from the Microsoft AI (MAI) family: MAI-Image-2.5 for image generation and editing, MAI-Transcribe-1.5 for high-speed speech-to-text, and MAI-Voice-2 for expressive text-to-speech. These models are designed to work together as a multimodal stack for developers building complex AI applications.
MAI-Transcribe-1.5 is designed for production-grade speed and accuracy. It can transcribe one hour of audio in less than 15 seconds. The model supports 43 languages and includes built-in support for domain-specific terminology, ranking #1 on the Artificial Analysis Accuracy x Speed Pareto frontier.
MAI-Voice-2 provides expressive text-to-speech capabilities across 15 languages. It allows for specific emotional control, enabling the generated voice to sound excited, embarrassed, or whispered. The model is also designed to maintain a stable speaker identity even when generating long-form content or adapting from a short voice sample.
No. Microsoft AI trains the MAI model family from the ground up on clean, licensed data. They do not use distillation, which is the process of training a smaller model to replicate the behavior of a larger teacher model. This approach is intended to ensure original performance and long-term self-sufficiency.
Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →
