For enterprises transcribing thousands of hours, the economics of transcription can matter just as much as the accuracy.
Microsoft has released MAI-Transcribe-2, a speech-recognition model designed to deliver faster and less expensive AI transcription. Microsoft says the model achieved a 5.2% average word error rate across 60 languages in its FLEURS benchmark testing, outperforming the Google, OpenAI, and ElevenLabs models included in its comparison.
The model is currently available through Microsoft Foundry and the MAI Playground.
For Microsoft partners, the model could lower the cost of building transcription services for contact centers, meeting platforms, media processing, and other voice AI applications.
How MAI-Transcribe-2 compares to other options
In April 2026, Microsoft released MAI-Transcribe-1, which supported 25 languages. This model was deprecated toward the end of last month.
In June 2026, the company then released MAI-Transcribe-1.5, which supports 43 languages. Unlike MAI-Transcribe-1, this model is still supported.
This month, Microsoft released MAI-Transcribe-2, which it said was 10 times faster than OpenAI’s GPT-Transcribe, seven times faster than ElevenLabs’ Scribe v2, and five times faster than Gemini 3.5 Transcribe in the company’s testing while delivering greater accuracy.
Microsoft’s results should be interpreted in the context of the company’s testing methodology. Although Microsoft reports a 5.2% average Word Error Rate (WER) on FLEURS, this does not mean that every real-world recording will have a 5.2% error rate.
Additionally, the 60-language average does not indicate equal performance across every language. Accuracy may vary by language, regional accent, audio quality, background noise, and speaking style.
Despite the variability in accuracy, Microsoft says that the model can handle noisy conditions, language switching and speaker diarization.
By comparison, Gemini 3.5 Transcribe supports more than 85 languages, while ElevenLabs Scribe v2 supports more than 90. Microsoft reported MAI-Transcribe-2 benchmark results across 60 languages, emphasizing transcription quality and batch-processing speed rather than the broadest language coverage.
The enterprise takeaway
MAI-Transcribe-2 combines multilingual transcription with introductory pricing intended to reduce the cost of large batch-processing workloads.
Microsoft is offering introductory pricing of $0.10 per audio hour through the end of 2026. The company compares that rate with prices as high as $0.36 per audio hour for competing transcription models, although costs vary by provider and service configuration.
The introductory rate gives partners and enterprise customers a lower-cost way to test Microsoft’s performance claims on workloads such as customer-service calls, meetings, interviews, and media archives.
For now, the immediate opportunity is to test whether MAI-Transcribe-2 can lower the cost of transcription-heavy services without sacrificing accuracy across customers’ languages and recording environments. The introductory price reduces the cost of evaluation, but partners should run their own representative tests before the offer expires at the end of 2026.