Speech

    FRFranceVoxtral Transcribe

    Mistral AI · France · EU / EEA

    Accurate multilingual speech-to-text — meetings, calls and interviews with speaker labels (diarization), custom vocabulary and real-time streaming, across 50 languages incl. all official EU languages. Among the most accurate transcription models on the independent AA benchmark.

    EU company
    EU data residency
    Open weights
    Compliance: Very high
    Open Voxtral Transcribe

    Classification

    Jurisdiction tier
    EU / EEA — EU member state or EEA (Norway, Iceland, Liechtenstein) — GDPR applies directly.
    Caveat
    Speech is a phase-2 category — listed in data, not in the launch picker. The Voxtral 3 models (Oct 2026) are in Public Preview and API-only unless you arrange self-deployment with Mistral; the open-weight option is the earlier Voxtral Mini 4B Realtime (13 languages). For Danish-only work, compare with Hviske.
    Compliance level
    Very high — Derived from the verified classification data on this site: Very high = EU/EEA jurisdiction + documented EU data residency + self-hostable option. High = EU/EEA + EU data residency. Medium = adequacy country + residency, or EU/EEA without documented residency. Each underlying claim carries its own source.
    Tags
    phase2
    api
    preview
    What the badges mean

    EU company — Vendor headquartered in Europe (state country + tier).

    EU data residency — EU data-residency / processing option available.

    Open weights — Open weights / self-hostable.

    Why these rules
    Last verified
    2026-10-25

    Sources

    AI Capability Snapshot

    AI Capability Index — normalised so the strongest model scores 100 (anchor: Claude Opus 5.5 (max effort)). How this is measured

    Capability indexNot yet scored
    Coding subscoreNot yet scored

    Speech recognition is not covered by the text capability index. On the independent Artificial Analysis speech-to-text benchmark (lower is better) Voxtral Small Transcribe 3 ranks 3rd at 2.1%.

    Published benchmarks

    Vendor-published code benchmarks. These are not directly comparable to the two normalised scores above.

    Scored on
    Not scored on the text capability index: speech recognition is measured by word error rate, not the AA Intelligence Index (see published benchmarks).
    Price level
    Low
    Retrieved
    2026-10-25

    Reference models (non-EU)

    For scale only — these models are not European and are not listed in the directory.

    • Claude Opus 5.5 · Anthropic (US)100 / 100
    • Claude Fable 5.1 · Anthropic (US)91 / 100
    • Claude Opus 5 · Anthropic (US)88 / 100
    • GPT-5.6 Sol · OpenAI (US)81 / 100
    • Gemini 3.8 Flash · Google (US)71 / 100

    Independent, informational guide — not legal, security or compliance advice. Classifications describe company location, GDPR transfer status, data-residency options and licensing as of the date shown, from public sources. Verify data-processing terms with the provider before use, especially for personal or sensitive data. No affiliation with any vendor.

    Back to the directoryFlag as outdated