Rare find

Voxtral. Turn speech into text using a small AI model that runs on your computer.

github.com/antirez/voxtral.c

Problem

The full read on this project is a sign-in away — what it solves, who it serves, and where it sits in the market.

Audience

The full read on this project is a sign-in away — what it solves, who it serves, and where it sits in the market.

Market

The full read on this project is a sign-in away — what it solves, who it serves, and where it sits in the market.

Pricing

The full read on this project is a sign-in away — what it solves, who it serves, and where it sits in the market.

Vaya's read on this project

Problem, audience, market, and the verdict — sign in to see it.

Updates

February 2026

README: update --monitor emoji lists. Feb 15
streaming: robust continuous stdin with full-state recovery Feb 15
Fix WAV parsing for piped ffmpeg output Feb 14
MPS speedups: various optimizations Feb 14
Compact adapter buffer by reclaiming consumed tokens Feb 8
Add --from-mic live microphone transcription and vox_stream_flush() API Feb 7
Merge pull request #2 from 3podi/main Feb 6
Document -I flag, latency/efficiency tradeoff, and updated benchmarks Feb 6
Fix bandwidth spec: M3 Max 128GB is 400 GB/s, not 300 GB/s Feb 6
SPEED.md broader credits rules, specify baseline hardware Feb 6
keep last DEC_WINDOW in kv cache Feb 6
remove for loop attn_mask Feb 6
Pre-warm encoder + prefill MPS matmul ops at load time Feb 6
Update SPEED.md with Q-tiled attention results (Attempt 14) Feb 6
Q-tiled encoder attention: BQ=8 queries per threadgroup (encoder 23% … Feb 6
Merged decoder prefill weights: QKV 3→1 + FFN 2→1 matmul (prefill ~33… Feb 6
Merged encoder weights: QKV 3→1 + FFN w1+w3 2→1 matmul (encoder ~310→… Feb 6
Add testing instructions to SPEED.md Feb 6
Add regression test: runtest.sh and make test target Feb 6
Fix encoder attention q_offset after KV cache compaction Feb 6