Qwen3-ASR 0.6B vs Fun-ASR-MLT-Nano-2512
Two compact multilingual models with strong Chinese coverage. Qwen3-ASR 0.6B ranks higher in 5 of the matcher's seven common setups (with 2 tied), but the right choice still depends on the differences below.
Key differences
- Qwen3-ASR 0.6B lists 30 languages in our record, Fun-ASR-MLT-Nano-2512 31; 22 are shared.
- Qwen3-ASR 0.6B carries the higher relative accuracy rating (5/5 versus 4/5); confirm it on your own audio.
- Qwen3-ASR 0.6B is rated better for Apple Silicon (3/5 versus 2/5).
- The main documented runtime is qwen-asr package for Qwen3-ASR 0.6B and FunASR for Fun-ASR-MLT-Nano-2512.
Side by side
| Fact | Qwen3-ASR 0.6B | Fun-ASR-MLT-Nano-2512 |
|---|---|---|
| Provider · size | Qwen · 0.6B | FunAudioLLM · 800M |
| License | Apache-2.0 Permissive | Apache-2.0 Permissive |
| Languages | 30 languages + 22 Chinese dialects | 31 listed languages |
| Streaming | Via runtime | Via runtime |
| Timestamps | Separate aligner | Not documented |
| Speaker diarization | Separate pipeline | Separate pipeline |
| CPU · Apple · NVIDIA fit | 2/5 · 3/5 · 5/5 | 2/5 · 2/5 · 5/5 |
| Live · long recordings | 4/5 · 4/5 | 3/5 · 4/5 |
| Main runtime | qwen-asr package | FunASR |
Which ranks higher in common setups
Scores from the same deterministic matcher as the homepage, applied to all 26 reviewed models (rank out of 26). They are decision guidance, not accuracy claims.
| Setup | Qwen3-ASR 0.6B | Fun-ASR-MLT-Nano-2512 | Higher |
|---|---|---|---|
| Long recordings on an 8–12 GB NVIDIA GPU, balanced | 75/100 · #8 | 75/100 · #9 | Tie |
| Live captions or voice input (native streaming + timestamps, speed first) | 79/100 · #10 | 73/100 · #15 | Qwen3-ASR 0.6B |
| Subtitles with timing, accuracy first | 80/100 · #7 | 68/100 · #15 | Qwen3-ASR 0.6B |
| Meetings with speaker labels (diarization + timestamps) | 77/100 · #8 | 72/100 · #14 | Qwen3-ASR 0.6B |
| CPU-only machine, low memory use | 58/100 · #17 | 58/100 · #18 | Tie |
| Apple Silicon with 8–16 GB unified memory, long recordings | 66/100 · #16 | 61/100 · #20 | Qwen3-ASR 0.6B |
| Japanese live transcription (native streaming, speed first) | 77/100 · #4 | 73/100 · #8 | Qwen3-ASR 0.6B |
Language coverage
Both list 22 languages: Arabic, Cantonese, Czech, Danish, Dutch, English, Filipino, Finnish, Greek, Hindi, Hungarian, Indonesian and 10 more. Only Qwen3-ASR 0.6B lists French, German, Italian, Macedonian, Persian, Russian, Spanish and Turkish. Only Fun-ASR-MLT-Nano-2512 lists Bulgarian, Croatian, Estonian, Irish, Latvian, Lithuanian, Maltese, Slovak and Slovenian.
How to run each
- Qwen3-ASR 0.6B
- qwen-asr package, vLLM, Docker image qwenllm/qwen3-asr.
pip install -U qwen-asr - Fun-ASR-MLT-Nano-2512
- FunASR.
python -m pip install -U "funasr>=1.4.1"The model card's example sets the device to a CUDA GPU or the CPU; no memory figure is published.
Runtimes and figures from the official sources checked on November 2, 2026; details on each Qwen3-ASR 0.6B and Fun-ASR-MLT-Nano-2512 profile.
Common questions
Is Qwen3-ASR 0.6B better than Fun-ASR-MLT-Nano-2512?
It depends on the job. Qwen3-ASR 0.6B ranks higher in 5 of the matcher's seven common setups (with 2 tied), but the right choice still depends on the differences below. Benchmark both on your own audio before deciding.
Which is lighter to run, Qwen3-ASR 0.6B or Fun-ASR-MLT-Nano-2512?
They carry the same memory rating (3/5). See the official guidance quoted in the run section above.
Which supports more languages?
Fun-ASR-MLT-Nano-2512 lists 31 languages against 30. Only Qwen3-ASR 0.6B lists French, German, Italian, Macedonian, Persian, Russian, Spanish and Turkish. Only Fun-ASR-MLT-Nano-2512 lists Bulgarian, Croatian, Estonian, Irish, Latvian, Lithuanian, Maltese, Slovak and 1 more.
Go further
- Add a third model in the interactive comparison
- Benchmark both on your own audio
- All model comparisons
Generated from the reviewed model database (latest review November 3, 2026). Differences are taken from primary sources; matcher scores are Speech Model Match's editorial decision scores.