Voxtral
Open-weight speech understanding: transcription, audio QA, and voice function-calling.
About
Mistral's first audio family, released July 15, 2025, comes in two Apache-2.0 sizes: Voxtral Small (24B, built on Mistral Small 3.1) and Voxtral Mini (3B, built on Ministral 3B). Both take a 32K-token context — about 30 minutes of audio for transcription, 40 for understanding — and go beyond transcription into audio question-answering, summarization, translation, and voice-triggered function calling. Mistral reports the Small beats Whisper large-v3 on FLEURS word error rate; that is vendor-reported, so test on your own audio. The Mini Transcribe API starts at $0.001 per audio minute. Honest caveats: the 24B needs around 55GB of VRAM, so serious self-hosting is not laptop-grade.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Vendor-reported: Voxtral Small beats Whisper large-v3 on FLEURS WER.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: July 15, 2025 release verified via multiple press reports; Apache 2.0 verified.; Two sizes (24B/3B), 32K context (~30 min audio), $0.001/min API pricing verified.; Vendor-reported FLEURS win over Whisper labeled as vendor-reported with advice to test own audio.; 55GB VRAM self-hosting caveat included..