- Mean of LibriSpeech test-clean 1.28 / test-other 2.42 1.85
- Fleurs (zh | en) 2.69 | 4.44 2.69
- AISHELL-1 0.6
- AISHELL-2 ios 2.56
- Mean of WenetSpeech test-meeting 6.28 / test-net 5.37 5.83
- Mean of MMAU music 61.68 / sound 73.27 / speech 60.66 65.2
- ClothoAQA test (dev 73.18) 71.2
- Mean of OpenAudioBench AlpacaEval 75.73 / Llama Questions 79.33 / Reasoning QA 58.02 / TriviaQA 62.10 / Web Questions 70.20 69.1
- VoiceBench Avg (AlpacaEval 4.46, CommonEval 3.97, SD-QA 63.12, MMSU 62.17, OpenBookQA 83.52, IFEval 61.10, AdvBench 100.00) 76.9
Kimi-Audio-7B-Instruct
Kimi-Audio-7B-Instruct places in the top three on 2 benchmarks.
15
Benchmarks with a score
0
Ranked first, out of five or more models
2
In the top three
Moonshot AI Apr 2025
Kimi-Audio-7B-Instruct
Open weights Small
No Epoch Capabilities Index score yet
- MMAU 65.2% #13 /15
- Audio MC 13.7% #17 /18
- MMAU-Pro 48.3% #8 /10
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AISHELL-2 | 2.56% | #1 of 3 | Lab model card | |
| MMSU | 59.3% | #2 of 5 | Official leaderboard | |
| MMAU v05.15.25 | 68.2% | #3 of 6 | Official leaderboard | |
| AISHELL-1 | 0.60% | #2 of 4 | Lab model card | |
| Kimi-ASR Internal Testset Subset 1 | 1.42% | #1 of 2 | Official leaderboard | |
| Kimi-ASR Internal Testset Subset 2 | 2.44% | #1 of 2 | Official leaderboard | |
| MMAU-Pro | 48.3% | #8 of 10 | Official leaderboard | |
| MMAU | 65.2% | #13 of 15 | Lab model card | |
| VoiceBench | 76.9% | #11 of 12 | Lab model card | |
| Audio MC | 13.7% | #17 of 18 | Official leaderboard | |
| Clotho-AQA | 71.2% | #3 of 3 | Lab model card | |
| SpotSoundBench | 3.20% | #3 of 3 | Official leaderboard | |
| OpenAudioBench | 69.1% | #2 of 2 | Lab model card | |
| WildSpeech-Bench | 4.54 | #2 of 2 | Official leaderboard | |
| Kimi-Audio speech conversation evaluation | 3.90 | #1 of 1 | Lab model card |
Its lab's numbers next to 2 leaderboards
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| VoiceBench | 76.9% VoiceBench Avg (AlpacaEval 4.46, CommonEval 3.97, SD-QA 63.12, MMSU 62.17, OpenBookQA 83.52, IFEval 61.10, AdvBench 100.00) | 76.9% Leaderboard | Same |
| Kimi-Audio speech conversation evaluation | 3.90 Speech conversation, human ratings on a 1 to 5 scale, Avg (Table 7). Speed Control 4.30, Accent Control 3.45, Emotion Control 4.27, Empathy 3.39, Style Control 4.09 | 3.90 Leaderboard | Same |
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.
Its system card
Every result in the longer documents the lab published with it. 2 of them aren't in the launch post.
Kimi-Audio Technical Report Apr 25, 2025