ModelMicrosoftConfirmed
Story 8 of 13 in the Tuesday 10 November editionMicrosoft AI introduces MAI Transcribe 2 speech model at $0.10 per audio hour
Supporting 60 languages across batch and 128 ms streaming endpoints, MAI Transcribe 2 processes an hour of audio in 10 seconds with built-in diarisation and posts a 5.2% average word error rate on FLEURS.
Edited by Ashutosh Shrivastava
11 Nov 2026, 09:20 IST · 3 min read
Microsoft AI introduced MAI Transcribe 2 and MAI Transcribe 2 Streaming across 60 languages on 10 November 2026, alongside a personal browser playground supporting MP3 and WAV uploads up to 25 MB and 60 minutes. Batch MAI Transcribe 2 includes built-in speaker diarisation, word-level timestamps, multilingual code-switching, and domain keyword biasing at a promotional API rate of $0.10 per audio hour through the end of 2026, while MAI Transcribe 2 Streaming targets live voice agents at $0.54 per audio hour without diarisation, word timestamps, or keyword biasing.
Throughput and 60-language FLEURS accuracy
Microsoft’s benchmark charts are rendered as interactive HTML visualisations on the model page rather than downloadable images: across all 60 FLEURS languages, MAI Transcribe 2 records a 5.2% average word error rate in both forced and automatic language-detection modes, ahead of Gemini 3.1 Pro (5.3% and 5.8%), Gemini 3.5 Transcribe (5.9% and 8.6%), ElevenLabs Scribe v2 (6.2% and 6.5%), GPT-Transcribe (10.4% and 10.6%), and Whisper v3-large (22.8% and 23.5%).
Key numbers
- $0.10 per audio hour promotional batch API price through 2026 ($0.54 per hour on MAI Transcribe 2 Streaming)
- 10 seconds to transcribe 1 hour of audio (403.6 speed factor, 10 times faster than GPT-Transcribe and 5 times faster than Gemini 3.5 Transcribe)
- 128 ms average model inference latency on MAI Transcribe 2 Streaming
- 3.4% average word error rate on top-25 FLEURS languages and 2.0% on the Artificial Analysis WER Index
Technical details
- Endpoint variants and rates
- MAI Transcribe 2 batch ($0.10 per audio hour promo through 2026) and MAI Transcribe 2 Streaming ($0.54 per hour)
- Throughput and latency
- 1 hour transcribed in 10 seconds (403.6 speed factor) on batch; 128 ms inference latency on Streaming
- Language and WER benchmarks
- 60 languages at 5.2% FLEURS average WER (3.4% top-25 FLEURS, 2.0% Artificial Analysis WER Index)
- Batch-exclusive features
- Built-in speaker diarisation, word-level timestamps, code-switching and domain keyword biasing
- MAI Playground limits
- MP3 or WAV up to 25 MB and 60 minutes (Clean or Verbatim mode, age 18 and over personal non-commercial use)
Release details & context
Batch-versus-streaming feature split and MAI Playground rules
On the MAI Transcribe 2 model page, Microsoft AI places MAI Transcribe 2 at #1 across both multilingual FLEURS and DAI-ASR-I18N, expanding its first-party model catalogue following MAI-Image-2.7 and Copilot Harness and Foundry updates.
Developers choosing between the two endpoints face an explicit capability split: batch MAI Transcribe 2 ($0.10 per audio hour through 2026) bundles speaker diarisation, word-level timestamps and custom domain keyword biasing, whereas the 128 ms MAI Transcribe 2 Streaming endpoint ($0.54 per audio hour) omits those three post-processing features for low-latency voice turn-taking. Browser testing in the MAI Playground offers Clean and Verbatim transcript toggles on microphone recordings or single MP3 and WAV files up to 25 MB and 60 minutes, gated behind Microsoft sign-in and age 18 confirmation for personal, non-commercial use only.
Common questions
Does MAI Transcribe 2 Streaming support speaker diarisation and keyword biasing?
No, speaker diarisation, word-level timestamps and domain keyword biasing are exclusive to the batch MAI Transcribe 2 endpoint ($0.10 per audio hour through 2026) and omitted from MAI Transcribe 2 Streaming ($0.54 per hour).
What file limits and usage rules apply to the browser MAI Playground?
Authenticated users aged 18 and over can record live audio or upload a single MP3 or WAV file up to 25 MB and 60 minutes in Clean or Verbatim format strictly for personal, non-commercial testing.
Sources
- PrimaryMAI Transcribe 2 official model pagemicrosoft.ai
- On XMustafa Suleyman (@mustafasuleyman) launch post on Xx.com
Microsoft release timeline (7 launches tracked)
- Microsoft adds agent containment controls to Project Perception
Microsoft’s Project Perception can now limit an agent as its risk score changes.
- Microsoft AI publishes MAI-DxO sequential diagnosis study in Science and open-sources SDBench
Microsoft AI published its MAI-DxO study in Science, reaching 83% accuracy on 304 NEJM cases (versus 22% closed-book and 36% open-book for physicians).
- Microsoft launches MAI-Image-2.7 in preview and makes MAI-Image-2.6 generally available
Microsoft introduced MAI-Image-2.7 in preview and brought MAI-Image-2.6 and 2.6-Flash to general availability in Azure AI Foundry.
- Microsoft brings the multi-model Copilot harness to Microsoft Foundry
Microsoft exposes the Rust-based Copilot harness in Foundry as an SDK and managed service supporting 20+ models, MCP tools, and BYOK deployments.