Voxtral Models by MistralAI
Last refreshed 2026-10-12. Next refresh: weekly.
Details
Capabilities
About
Voxtral is Mistral AI’s open speech-understanding model family for transcription and audio comprehension, released as a larger production-oriented variant and a smaller Mini variant for local and edge use. Beyond speech-to-text, the models support long-form audio context, built-in Q&A and summarization over audio, multilingual recognition, and function calling from voice, while keeping text ability from their Mistral Small–class backbone. Mistral publishes weights for download and also offers API access, including a transcribe-optimized Mini route.
Decision facts
- Best fit
- speech recognitiontext to speechaudio
- Capability starting point
- Voxtral Mini Transcribe 2 with 32K context and multimodal inputs
- Lowest tracked input
- Mistral Voxtral Mini 3B 2507 · $0.040/1M · AWS Bedrock
- Closest related family
- Devstral
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs speech recognition, 4.4B parameters, and multimodal inputs.
Use when the workload needs text to speech, multimodal inputs, and audio.
Use when the workload needs speech recognition, 32K context, and multimodal inputs.
Use when the workload needs audio and 3B parameters.
Use when the workload needs audio and 24B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Voxtral Mini 4B Realtime Arabic | Use when the workload needs speech recognition, 4.4B parameters, and multimodal inputs. | 2026-10 | speech recognition4.4B parametersmultimodal inputs | Current |
| Voxtral TTS | Use when the workload needs text to speech, multimodal inputs, and audio. | 2026-03 | text to speechmultimodal inputsaudio | Current |
| Voxtral Mini Transcribe 2 | Use when the workload needs speech recognition, 32K context, and multimodal inputs. | 2026-02 | speech recognition32K contextmultimodal inputs | Current |
| Mistral Voxtral Mini 3B 2507 | Use when the workload needs audio and 3B parameters. | 2025-07 | audio3B parameters | Current |
| Mistral Voxtral Small 24B 2507 | Use when the workload needs audio and 24B parameters. | 2025-07 | audio24B parameters | Current |
Release Timeline
4 release groupsSpecifications(5 models)
| Model | Released | Context | Parameters | Multimodal |
|---|---|---|---|---|
| Voxtral Mini 4B Realtime Arabic | 2026-10 | — | 4.4B | Yes |
| Voxtral TTS | 2026-03 | — | — | Yes |
| Voxtral Mini Transcribe 2 | 2026-02 | 32K | — | Yes |
| Mistral Voxtral Mini 3B 2507 | 2025-07 | — | 3B | No |
| Mistral Voxtral Small 24B 2507 | 2025-07 | — | 24B | No |
Available From(2 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| Mistral Voxtral Mini 3B 2507 | AWS Bedrock | $0.04 | $0.04 | Serverless |
| Mistral Voxtral Small 24B 2507 | AWS Bedrock | $0.1 | $0.3 | Serverless |
| Mistral Voxtral Small 24B 2507 | Mistral AI Studio | $0.1 | $0.4 | Serverless |
