Microsoft's MAI-Transcribe-1 achieves lowest word error rate on FLEURS, costs $0.36/audio hour
Microsoft has released MAI-Transcribe-1, a speech-to-text model that achieves the lowest word error rate on the FLEURS benchmark across 25 languages, outperforming Whisper-large-V3, GPT-Transcribe, and Gemini 3.1 Flash-Lite. The model runs 2.5 times faster than Microsoft's previous Azure Fast offering and costs $0.36 per audio hour.
Microsoft's MAI-Transcribe-1 Achieves Lowest Word Error Rate on FLEURS Benchmark
Microsoft has introduced MAI-Transcribe-1, a multilingual speech-to-text model supporting 25 languages that outperforms competing transcription systems on the FLEURS benchmark.
Performance and Capabilities
MAI-Transcribe-1 achieves the lowest word error rate among tested models, beating Scribe v2, Whisper-large-V3, GPT-Transcribe, and Gemini 3.1 Flash-Lite across the FLEURS evaluation suite. Microsoft says the model is optimized for challenging recording conditions, including background noise, poor audio quality, and overlapping speech.
The model delivers 2.5x faster inference than Microsoft's previous Azure Fast transcription offering. When combined with MAI-Voice-1 (Microsoft's text-to-speech model) and a language model, MAI-Transcribe-1 can power voice agents, according to Microsoft.
Pricing and Availability
MAI-Transcribe-1 is priced at $0.36 per audio hour. The model is rolling out across Copilot Voice and Microsoft Teams. Developers can access it through a public preview on Microsoft Foundry and the Microsoft AI Playground.
Market Context
The release comes as open-source alternatives gain traction. Cohere and Mistral recently released open-source speech-to-text models that perform at comparable quality levels, offering cost-free deployment options for organizations willing to handle self-hosting infrastructure.
What This Means
MAI-Transcribe-1 positions Microsoft competitively in speech recognition, addressing both accuracy and speed requirements for enterprise voice applications. The $0.36/hour pricing sits in the mid-market range for commercial transcription APIs. However, the emergence of capable open-source alternatives means Microsoft must justify the API model through deployment convenience and integration with Copilot and Teams ecosystems rather than technology superiority alone. The 2.5x speed improvement over Azure Fast suggests meaningful optimization work, relevant for real-time voice agent applications.
Related Articles
Microsoft to Kill Excel's COPILOT() Function on September 14, 2026
Microsoft will shut down Excel's COPILOT() worksheet function on September 14, 2026, roughly a year after its preview launch. The company says the Copilot side pane already covers the same capabilities, so a planned 2027 general availability release has been scrapped.
DeepSeek Releases Experimental V4-Flash-Vision-Exp, Claims Near-Parity With Opus 4.8 on Agent Benchmarks
DeepSeek has released V4-Flash-Vision-Exp, an experimental multimodal extension of V4-Flash that adds image understanding while preserving text reasoning capabilities. The company claims the model approaches or beats Anthropic's Opus 4.8 on its internal multimodal agent benchmarks.
GitHub Copilot Now Works as a Shared Agent Inside Microsoft Teams Channels
GitHub has launched a feature that lets teams mention @GitHub in Microsoft Teams channels, threads, or DMs to start a shared Copilot agent session visible to all participants. The update turns individual Copilot agent work into a collaborative, team-visible process directly inside Teams.
DeepSeek Releases V4 Flash Vision Exp, an Experimental Multimodal MoE Model with 1M Context
DeepSeek has released V4 Flash Vision Exp, an experimental vision-enabled variant of DeepSeek V4 Flash 0731 that adds image understanding while matching the base model's text performance. The sparse mixture-of-experts model uses 13B active parameters out of 284B total and supports a 1M token context window.
Comments
Loading...