Zoom Scribe V2.1 Pro Live tops streaming STT accuracy at 2.4% WER
ArtificialAnlys · x · 2026-11-20
Artificial Analysis independently benchmarked Zoom's new streaming speech-to-text model, Zoom Scribe V2.1 Pro Live, which takes the #1 spot on the AA-WER Streaming leaderboard among 40+ models with a 2.4% final transcript word error rate, just ahead of Microsoft's MAI-Transcribe-2-Streaming at 2.5%.
The accuracy win comes with a latency tradeoff: Zoom returns the final transcript 0.66s after end of speech versus 0.13s for the Microsoft model, and its first partial transcript is both less accurate (5.0% vs 2.5% WER) and slower (0.55s vs 0.12s).
Related event: Zoom Scribe Tops Streaming ASR Benchmark at 2.4% WER, but Lags in Latency(4 posts)→
More from Models
- OpenAI: keep conversation prefix unchanged when using GPT-6.2 in Responses API — OpenAIDevs · 2026-11-20
- Engineer's first take on Hy4-Flash: clean code out of the box, hours of heavy use cost $0.10 — crusaderky · 2026-11-20
- Zoom Scribe pricing: $0.40/hour, between Microsoft and ElevenLabs — ArtificialAnlys · 2026-11-20
- Gemini API adds tool search preview: defer tool loading so agents skip sending full tool libraries — rseroter · 2026-11-20
- scaling01 revises Google verdict: now just 1-2 months behind frontier, not 3-4 — scaling01 · 2026-11-20
- Cognition: Gemini 4 Neon scores 46.2% on FrontierCode 2.0, near Opus 5.5 at 1/3 cost — scaling01 · 2026-11-20