Sauti / xai
Grok STT
xAI's only speech-to-text model
WER
9.8%
95% CI 8.9%–10.9%
Rank
#1
of 12 measured
Latency
1.5s
mean, per clip
Cost
$0.10
per audio hour
6 other models are a statistical tie with this one — their confidence intervals overlap, so the ordering between them is not something this data can settle: nova-3, gpt-transcribe, Whisper large-v3 (self-hosted), universal-2, universal-3-5-pro, Whisper small (self-hosted).
What it is
A batch transcription endpoint on the xAI API. xAI does not document a model identifier for it, and publishes one rate for the whole service rather than per model.
Pick it when
It topped our English conversational board while carrying the lowest paid price on it, which is an unusual combination and the reason it is worth a look despite being the least-discussed model here.
Think twice when
Coverage is roughly 25 languages, so it is an English-first choice. It is also the model we can say least about: no published id and no per-language figures. And read the tie list above before treating its position as a win — much of the board overlaps it.
Assessment written against the board of 2026-08-08. The figures above are read live from the current published snapshot (2026-08-08), so if those dates differ, trust the figures.
What the vendor publishes
xai publishes no stated word error rate figure for this model on any public benchmark. That absence is itself the finding. Source (read 2026-08-15).xAI documents pricing for the speech-to-text service and no performance metrics at all — no WER, no benchmark, no language-level accuracy claim.
How this was measured
250 clips (2.9 hours) of unscripted conversational English from People’s Speech (MLCommons, CC-BY), human-transcribed. Scored on a held-out split (N=77) never used to tune anything, through one fixed text normalizer, with bootstrap 95% confidence intervals. One corpus, one language, one modality: batch pre-recorded. It does not tell you how this model handles your accents, your domain vocabulary, or streaming.
Cost is xai’s own published pre-recorded pay-as-you-go list price, read from their pricing page on 2026-08-08. Streaming and committed-volume rates differ.Published as $0.10/hr REST. Streaming is $0.20/hr. xAI does not document a model id, so this is provider-wide.
Other models we measured
- Deepgram Nova-311.5%
- OpenAI gpt-transcribe11.8%
- OpenAI gpt-4o-mini-transcribe12.8%
- OpenAI gpt-4o-transcribe14.7%
- OpenAI whisper-113.5%
- AssemblyAI Universal-3.5 Pro12.3%
- AssemblyAI Universal-212.1%
- ElevenLabs Scribe v213.6%
- ElevenLabs Scribe v113.7%
- Whisper large-v3 (self-hosted)11.9%
- Whisper small (self-hosted)12.3%