Gemini 3.5 Transcribe Hits 84x Realtime at $5/1,000 min, Ranks #5 on AA-WER
ArtificialAnlys · x · 2026-08-27
Google released Gemini 3.5 Transcribe, ranking #5 on Artificial Analysis's AA-WER benchmark for non-streaming transcription at 2.6% WER, alongside Gemini 3.5 Transcribe Live, which streams continuously via the Live API at 4.0% AA-WER with 0.40s latency after end of speech.
It processes 84 seconds of audio per second (84x realtime) at $5 per 1,000 minutes. Among top-five accuracy models it's faster than ElevenLabs Scribe v2 (55x, $3.67) but slower than Microsoft MAI-Transcribe-1.5 (191x) and Smallest AI Pulse Pro (275x, $4). The two offerings target pre-recorded audio (Interactions API) and live streaming (Live API) respectively.
Related event: Google Launches Gemini 3.5 Transcribe Speech-to-Text Model(29 posts)→
More from Models
- Claude auto-files a bug report about itself: TLDRs still too long after user asked for shorter — andrey_kurenkov · 2026-10-13
- Meta skipped the 'too dangerous to release' marketing for Muse Wonder — and could rival Google — haider1 · 2026-10-13
- Anthropic says new defenses cut distillation attack data harvesting by 90-99%, cites Zhipu's GLM-5.3 — AnthropicAI · 2026-10-13
- Delve Town's 'bee' model was further trained by Claude for ~24 hours — Sauers_ · 2026-10-13
- Muse Wonder disappoints on RareBench: Opus 5.5 leads at 52% recall — danielmckinn0n · 2026-10-13
- Chart sparks debate: Are Anthropic and OpenAI really months ahead of Meta and Google? — scaling01 · 2026-10-13