Voxtral Mini 4B Realtime
Ranked 8 of 25 speech recognition models by recommenders
Open-weights speech-to-text model under Apache 2.0 that people run locally for real-time dictation. The most common caveat is its size: a 2.5GB download and memory footprint, and it is too slow on some machines.
Open-weights speech-to-text model under Apache 2.0 that people run locally for real-time dictation. The most common caveat is its size: a 2.5GB download and memory footprint, and it is too slow on some machines.
24 people gave an opinion
- recommend it
- 17
- criticize it
- 7
- approval share
- 71%
Recommended in 3 of 13 months
- November 2025: not recommended
- December 2025: not recommended
- January 2026: not recommended
- February 2026: recommended
- March 2026: not recommended
- April 2026: recommended
- May 2026: not recommended
- June 2026: not recommended
- July 2026: not recommended
- August 2026: not recommended
- September 2026: not recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it doesn’t carry the badge
- 17 people recommend itNeeds 10 or more, met.
- 2.4 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 3 of 13 monthsNeeds 4 or more, not met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Real-time transcription on capable hardware
“the model is definitely real time (able to transcribe audio faster than the length of the audio)”
Comment from 12 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser” “it is realtime on my DGX Spark with an RTF of .416”
Comment from 12 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser” “I'm using mistral voxtral realtime daily – it's great.”
Comment from 29 Apr 2026 in “Mistral Medium 3.5” Open weights, transparent, on-premise use
“From a cybersecurity perspective, this project is impressive not just for performance, but for transparency.”
Comment from 10 Feb 2026 in “Pure C, CPU-only inference with Mistral Voxtral Realtime 4B speech to text model” “open-models running on-premise. Preferred by users and businesses.”
Comment from 10 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser” “The Apache 2.0 license on Realtime is the buried lede.”
Comment from 5 Feb 2026 in “Voxtral Transcribe 2” Used for daily dictation
“I built a macOS dictation app on top of Voxtral Realtime”
Comment from 25 Feb 2026 in “Show HN: Moonshine Open-Weights STT models – higher accuracy than WhisperLargev3” “I'm using mistral voxtral realtime daily – it's great.”
Comment from 29 Apr 2026 in “Mistral Medium 3.5” “Not perfect dictation, but close enough”
Comment from 10 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser” Causal encoder, accurate for dictation
“The thing that sold me on Voxtral Realtime over Whisper-based models for dictation is the causal encoder.”
Comment from 27 Feb 2026 in “Parakeet.cpp – Parakeet ASR inference in pure C++ with Metal GPU acceleration” “At 480ms delay it matches offline models on accuracy (FLEURS benchmark)”
Comment from 24 Feb 2026 in “Show HN: Localvoxtral – Local real-time dictation on macOS with streaming STT”
What people warn about
Each mark is one comment
Too slow for real-time use
“Given that it took 19.64 mins to transcribe the 11 second sample wav”
Comment from 10 Feb 2026 in “Pure C, CPU-only inference with Mistral Voxtral Realtime 4B speech to text model” “This isn't even close to realtime on M4 Max.”
Comment from 10 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser” “way too slow for anything close to realtime”
Comment from 11 Feb 2026 in “Pure C, CPU-only inference with Mistral Voxtral Realtime 4B speech to text model” - Comment from 10 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser”
- Comment from 10 Feb 2026 in “Pure C, CPU-only inference with Mistral Voxtral Realtime 4B speech to text model”
Large 2.5GB download and memory
“I don't want to live in a world where its normal for a website to download 2.5Gb in the background to run something”
Comment from 10 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser” “this is not yet practical for browser use due to it's (at least) 2.5GB memory footprint”
Comment from 10 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser” “it is too big for local inference”
Comment from 10 Feb 2026 in “Pure C, CPU-only inference with Mistral Voxtral Realtime 4B speech to text model” Fails to load or run on some machines
“init failed: Worker error: Uncaught RuntimeError: unreachable”
Comment from 10 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser” “seems broken on my machine (Firefox, Asahi Linux, M1 Pro)”
Comment from 10 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser” “okay... so I cannot get this to run on my mac”
Comment from 10 Feb 2026 in “Rust implementation of Mistral's Voxtral Mini 4B Realtime runs in your browser”
Other speech recognition models people recommend
The four most recommended other speech recognition models, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| OpenAI Whisper | 164 | 53 | 76% | 12 | — | Approved |
| NVIDIA Parakeet | 144 | 27 | 84% | 12 | — | Approved |
| Mistral Voxtral | 69 | 27 | 72% | 8 | — | Approved |
| Moonshine open-weights STT models | 37 | 13 | 74% | 7 | — | Approved |
| Voxtral Mini 4B RealtimeThis page | 17 | 7 | 71% | 3 | — | — |
What people said
27 of the 30 comments that mention it, newest first. Each links to the original comment.
“Their previous generation open weight STT model was pretty good.”
“I'm using mistral voxtral realtime daily – it's great.”
“The thing that sold me on Voxtral Realtime over Whisper-based models for dictation is the causal encoder.”
“I built a macOS dictation app on top of Voxtral Realtime”
“At 480ms delay it matches offline models on accuracy (FLEURS benchmark)”
“it is realtime on my DGX Spark with an RTF of .416”
“the model is definitely real time (able to transcribe audio faster than the length of the audio)”
“way too slow for anything close to realtime”
“It starts to load, but hangs or is too slow.”
“That's great; should really increase the places where it's useful”
“From a cybersecurity perspective, this project is impressive not just for performance, but for transparency.”
“it is too big for local inference”
“this is not yet practical for browser use due to it's (at least) 2.5GB memory footprint”
“okay... so I cannot get this to run on my mac”
“I don't want to live in a world where its normal for a website to download 2.5Gb in the background to run something”
“Given that it took 19.64 mins to transcribe the 11 second sample wav”
“This isn't even close to realtime on M4 Max.”
“open-models running on-premise. Preferred by users and businesses.”
“I tried the demo and it looks like you have to click Mic, then record your audio, then click "Stop and transcribe" in order to see the result.”
“The Hugging Face server-side demo at https://huggingface.co/spaces/mistralai/Voxtral-Mini-Realtim... manages that”
“isn't even close to realtime”
“Not perfect dictation, but close enough”
“The second test was better, it detected english.”
“This was a breeze to install on Linux.”
“seems broken on my machine (Firefox, Asahi Linux, M1 Pro)”
“init failed: Worker error: Uncaught RuntimeError: unreachable”
“The Apache 2.0 license on Realtime is the buried lede.”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.