Voxtral Mini Realtime 2602
Voxtral Mini Realtime 2602 is a downloadable streaming transcription model. It supports 13 languages and ships in BF16 under Apache 2.0. Its serverless version is deprecated in favor of Realtime 3.
Deprecated API; check access details
Understand the streaming architecture
The model combines an approximately 3.4B language component with a 970M audio encoder. The audio encoder uses causal attention. Sliding-window attention in both parts supports continued streaming. The card lists a configurable delay from 240 milliseconds to 2.4 seconds.Voxtral Mini Realtime 2602 official weight card
Separate the deprecated API from weights
The model page records API deprecation on 31 October 2026. The official weight repository remains available. API access is subject to Mistral's retirement schedule; a published model identifier does not guarantee continued service after retirement.Voxtral Mini Realtime 2602 API lifecycle pageVoxtral Mini Realtime 2602 official weight card
Evaluate the complete task
Practical guidance. This release is useful for evaluating local live captions when a public weight license is required. Measure how transcript revisions affect a reader, as well as the final word error rate. Lower delay can change the information available when the model commits to a word.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Deprecated API; check access details |
| Access | API, Downloadable |
| Model identifiers | voxtral-mini-transcribe-realtime-2602Voxtral-Mini-4B-Realtime-2602 |
| API status | Deprecated on 31 October 2026Voxtral Mini Realtime 2602 API lifecycle page |
| Weight license | Apache 2.0Voxtral Mini Realtime 2602 official weight card |
| Weight repository | mistralai/Voxtral-Mini-4B-Realtime-2602Voxtral Mini Realtime 2602 official weight card |
Related models
- Voxtral Small 2507
Mistral AI · Voxtral
An Apache-licensed 24B audio-understanding model that answers questions about recordings and requests tools from spoken instructions.
Launch date API · Downloadable A streaming transcription model with turn detection, speaker diarization and configurable latency over WebSocket.
Launch date PreviewAPI