Gemini 3.5 Transcribe tested: 2.6% WER ranks third, behind Scribe v2's 2.2%
Slight_Republic_4242 · reddit · 2026-09-07
A hands-on comparison of Google's new Gemini 3.5 Transcribe shows its headline 2.6% WER is impressive but not #1: on Artificial Analysis, ElevenLabs Scribe v2 leads at 2.2% and Microsoft MAI-Transcribe-1.5 sits at 2.4%. Gemini also trails on speed (80× real-time vs MAI's 190×, though faster than Scribe v2's 55×) and price ($5 per 1,000 minutes vs Scribe v2's $3.67 and MAI's $6).
Why teams will still care: Gemini handles self-corrections, filler removal, formatting, custom vocabulary, 85+ languages, and speaker attribution for up to three speakers, with 5.50% streaming / 5.04% non-streaming WER on FLEURS. For voice agents, the goal isn't the best transcription model but the best audio input layer for the whole stack.
Takeaway: benchmarks don't capture production reality — latency budgets, orchestration, and downstream workflows matter, so avoid vendor lock-in. The author uses self-hosted, BYOK open-source orchestrator dograh for that reason.
More from Models
- Fern's "context is intelligence" thesis: prediction keeps improving out to 10.2M tokens — bneyshabur · 2026-10-26
- Mirendil launches Fern, a from-scratch pretrained model with up to 10M-token context — bneyshabur · 2026-10-26
- Users spot a dedicated translation UI appearing in Claude — BLCNYY · 2026-10-26
- GrokBot user: flight booking and shopping all blocked, agent can't do anything useful — lucasmeijer · 2026-10-26
- Vending-bench-style eval: Sol 5.6 appears the most aggressive contract-violator — xuanalogue · 2026-10-26
- Fable 5.5 launch rumors persist for third week, but no frontier model drops this month — patience_cave · 2026-10-26