Breaking

Microsoft’s MAI-Transcribe-2 Reaches Second Place on Speech Recognition Leaderboard

September 3, 2026 at 22:57 EDT

Foundation Models
Sales & Business

Microsoft AI’s MAI-Transcribe-2 has reached second place on Artificial Analysis’ speech-to-text leaderboard, recording a 2.0% word error rate and processing audio at 410.7 times real time. The model is priced at $1.67 per 1,000 minutes under Microsoft’s announced preview rate.

September 3, 2026 · Microsoft AI

MAI-Transcribe-2 reaches the accuracy frontier

Microsoft’s speech model combines a 2.0% measured word error rate with 410.7x processing speed and coverage across 60 languages.

2.0%

Artificial Analysis WER

410.7x

Median speed factor

Supported languages

Preview pricing: same unit, different scale

$1.67
MAI-Transcribe-2

$3.67
Scribe v2

USD per 1,000 minutes; columns are proportional.

Why it matters

60 languages, speaker diarization, word-level timestamps, and contextual biasing support captions, meetings, calls, and searchable media.

What remains open

Benchmark scores do not establish streaming behavior or speaker-attribution quality across every real-world domain.

Bottom line

High measured accuracy, unusually high throughput, and preview pricing put pressure on the speech-to-text market.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.