Primary use stated or inferred cautiously from official documentation.
M
MISTRAL / voxtral-mini-transcribe-realtime-3-0
Voxtral Mini Transcribe Realtime 3
Streaming audio transcriber presented by Mistral on October 20, 2026. The current guide lists 45 languages and dialects. It uses WebSocket, delivers incremental text, and adds turn-ending detection, speaker tags, and context_bias. It does not generate voice nor automatically share Small's offline capabilities. The versioned identifier allows distinguishing it from aliases that may change versions.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | voxtral-mini-transcribe-realtime-3-0 |
|---|---|
| Model type | Voice and audio |
| Access model | Paid proprietary |
| License | Proprietary |
| Deployment | Hosted API |
| Release | 20 OCT 2026 |
| Knowledge cutoff | Not published |
| Entrance | Audio |
| Exit | Text |
| Context window | Audio session via WebSocket |
| maximum output | Transcribed text |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Mistral API |
Live subtitles and voice interfaces that require incremental text. The configuration is sent before the first audio; modifying it afterwards produces an error. Latency settings accept integer values from 80 to 1,200 ms or 2,400 ms, not any intermediate value. These are model options, not a guarantee of total network and application latency.
Check reconnections, lost audio, and turn endings before automating actions. The 45 languages and dialects come from the specific guide, which is more recent than the general counts in the announcement. Do not confuse speaker detection with biometric identification. Audio may contain personal data: limit retention and access, inform individuals, and review errors in noisy environments.
03 / CAPABILITIES
What it can do
Orientation
Live subtitles and voice interfaces that require incremental text. The configuration is sent before the first audio; modifying it afterwards produces an error. Latency settings accept integer values from 80 to 1,200 ms or 2,400 ms, not any intermediate value. These are model options, not a guarantee of total network and application latency.
Context
Audio session via WebSocket · Transcribed text
Tools and integration
Not published
Access
Mistral API
Input modalities
Audio
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | 0.008 USD / min | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Included in the audio rate | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
Public rates in dollars per minute of audio or per million characters, depending on the variant; they are not token prices. Numerical fields per million tokens are left blank to avoid false comparisons. Confirm billing conditions and availability before integrating. Private enterprise deployment requires a specific agreement; no downloadable weights have been published for these variants.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
The announcement describes a FLEURS evaluation on a subset of 35 languages, with CER for Japanese, Korean, and Chinese. That evaluation set does not equate to the total declared language coverage. We have not verified per-language scores or reproduced the results. Compare accuracy, partial stability, turn detection, and end-to-end latency using the same recordings.
i Provider evaluation; numerical scores not verified by Inferama.
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Voxtral Mini Transcribe Realtime 3
Version added to Inferama's verified catalog.
ANALYSIS
Related analysis
07 / SOURCES