Microsoft's MAI-Transcribe-2 hits 2.0% WER at 411x real-time, priced at $1.67 per 1,000 minutes
ArtificialAnlys · x · 2026-09-03
Artificial Analysis benchmarks Microsoft AI's new MAI-Transcribe-2 speech model: 2.0% AA-WER (#2, behind Alibaba's Fun-Realtime-ASR-preview at 1.7%), 411x real-time speed (7x faster than ElevenLabs Scribe v2), and $1.67 per 1,000 minutes — less than half the price of Scribe v2 ($3.67) or Smallest AI Pulse Pro ($4). Language support expands from 43 to 60 languages, adding speaker diarization, word-level timestamps, and clean/verbatim styles. It tops the accuracy-speed Pareto frontier among high-accuracy models.
More from Models
- Codex, Grok and Claude All Go Down at the Same Time, and Devs Joke About the Robot Rebellion — iannuttall · 2026-09-03
- Meta's new Muse model suddenly outputs Chinese characters, sparking distillation rumors — zsakib_ · 2026-09-03
- K2-Horizon open-weight family (3.7B-375B, 512K ctx, Apache-2.0) gets day-0 vLLM support — HongyiWang10 · 2026-09-03
- Codex hit by widespread outage, OpenAI says it's investigating — DeryaTR_ · 2026-09-03
- With Codex and Claude Both Down, GLM and DeepSeek Held the Fort — omarsar0 · 2026-09-03
- OpenEvidence Launches Medical AI Model Family, Darwin Scores First-Ever 100% on MedQA — benxneo · 2026-09-03