CMI-RewardBench
Models judge musical quality and how well recordings follow requests. Some requests combine text, lyrics, and reference audio. Scores measure agreement with human listeners.
The CloudEar paper independently evaluates CMI-Test. The original paper compares three Qwen models and three Gemini models alongside specialist music judges.
- Released
- February 2026
- Kind
- Audio understanding and voice
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Prediction
- Tasks
- All public
- Measures
- Capability
- Subject
- Arts
- Language
- English
- Interaction
- Single turn
- Items from
- Written by crowdworkers
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Audio and speech
- Strongest model tested
-
Gemini 3 Pro #56 on the capability index
- Last model added
- November 2025
- Scores refreshed
- November 20th, 2026
SongEval-RM tops the leaderboard with 0.70 Gemini 3 Pro leads with 0.60 among models released by November 2025
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Gemini 3 Pro, at 0.60. This view shows only models we track, each at its best.
The newest model with a score here came out in November 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1 SongEval-RM 0.70
- 2 CMI-RM w/ f.t.: CMI-Pref 0.69
- 3
Gemini 3 Pro 0.60
- 4 PAM score 0.59
- 5 audiobox-CE 0.53
- 6 audiobox-CU 0.46
- 7 audiobox-PQ 0.46
- 8 Gemini2.5-pro (snapshot unspecified) 0.45
- 9 Qwen3-omni 0.42
- 10
Gemini 2.5 Flash Sep 2025 0.38
- 11 Omni-Reward 0.34
- 12 Qwen2.5-omni 0.28
- 13 audiobox-PC 0.25
- 14
Qwen2-Audio-7B-Instruct 0.15
Source: Official CMI-RewardBench leaderboard
- 1
Gemini 3 Pro 0.60
- 2
Gemini 2.5 Flash Sep 2025 0.38
- 3
Qwen2-Audio-7B-Instruct 0.15
Source: Official CMI-RewardBench leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | SongEval-RM | 0.70 | Official leaderboard | ||
| 2 | CMI-RM | 0.69 | w/ f.t.: CMI-Pref | Official leaderboard | |
| 3 | | 0.60 | Nov 2025 | Official leaderboard | |
| 4 | PAM score | 0.59 | Official leaderboard | ||
| 5 | audiobox-CE | 0.53 | Official leaderboard | ||
| 6 | audiobox-CU | 0.46 | Official leaderboard | ||
| 7 | audiobox-PQ | 0.46 | Official leaderboard | ||
| 8 | Gemini2.5-pro (snapshot unspecified) | 0.45 | Official leaderboard | ||
| 9 | Qwen3-omni | 0.42 | Official leaderboard | ||
| 10 | | 0.38 | Sep 2025 | Official leaderboard | |
| 11 | Omni-Reward | 0.34 | Official leaderboard | ||
| 12 | Qwen2.5-omni | 0.28 | Official leaderboard | ||
| 13 | audiobox-PC | 0.25 | Official leaderboard | ||
| 14 | | 0.15 | Aug 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 0.60 | Nov 2025 | Official leaderboard | |
| 2 | | 0.38 | Sep 2025 | Official leaderboard | |
| 3 | | 0.15 | Aug 2024 | Official leaderboard |
Rankings by part
We read 11 parts from the benchmark’s own leaderboard in 2 groups: musicality and alignment. Each ranks models on one part of the test. Gemini 3 Pro leads 10 of them, and Gemini 2.5 Flash leads the other one. Arrows show how far a model moved from its overall rank among the models we track.
PAM LCC
3 models we track have a score here.
- 1
Gemini 3 Pro 0.60
- 2
Gemini 2.5 Flash Sep 2025 0.38
- 3
Qwen2-Audio-7B-Instruct 0.15
MusicEval LCC
3 models we track have a score here.
- 1
Gemini 3 Pro 0.60
- 2
Gemini 2.5 Flash Sep 2025 0.42
- 3
Qwen2-Audio-7B-Instruct 0.15
Music Arena accuracy
3 models we track have a score here.
- 1
Gemini 3 Pro 68.8%
- 2
Gemini 2.5 Flash Sep 2025 64.1%
- 3
Qwen2-Audio-7B-Instruct 5.99%
CMI-Pref accuracy
3 models we track have a score here.
- 1
Gemini 3 Pro 65.8%
- 2
Gemini 2.5 Flash Sep 2025 64.2%
- 3
Qwen2-Audio-7B-Instruct 8.60%
PAM alignment LCC
3 models we track have a score here.
- 1
Gemini 3 Pro 0.52
- 2
Gemini 2.5 Flash Sep 2025 0.37
- 3
Qwen2-Audio-7B-Instruct -0.02
Text–Music
3 models we track have a score here.
- 1
Gemini 3 Pro 67.2%
- 2
Gemini 2.5 Flash Sep 2025 65.6%
- 3
Qwen2-Audio-7B-Instruct 0.80%
Text and lyrics
3 models we track have a score here.
- 1
Gemini 3 Pro 60.8%
- 2
Gemini 2.5 Flash Sep 2025 56%
- 3
Qwen2-Audio-7B-Instruct 2.40%
Text and audio
3 models we track have a score here.
- 1
Gemini 2.5 Flash Sep 2025 ↑1 69.6%
- 2
Gemini 3 Pro ↓1 68.8%
- 3
Qwen2-Audio-7B-Instruct 8.80%
Text, lyrics and audio
3 models we track have a score here.
- 1
Gemini 3 Pro 64.8%
- 2
Gemini 2.5 Flash Sep 2025 54.4%
- 3
Qwen2-Audio-7B-Instruct 14.4%
Without audio
3 models we track have a score here.
- 1
Gemini 3 Pro 64%
- 2
Gemini 2.5 Flash Sep 2025 60.8%
- 3
Qwen2-Audio-7B-Instruct 1.60%
With audio
3 models we track have a score here.
- 1
Gemini 3 Pro 66.8%
- 2
Gemini 2.5 Flash Sep 2025 62%
- 3
Qwen2-Audio-7B-Instruct 11.6%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official CMI-RewardBench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official CMI-RewardBench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Audio and video
- Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy.
Qwen 3.8 Max 88.7%
- LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization.
Gemini 4 Argon 91.7%
- MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy.
Qwen 3.8 Max 82.4%
- MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy.
Qwen 3.8 Max 90.8%
- FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better.
Phi-4-multimodal 4.00%
- VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules.
Qwen3.5-Omni-Plus 93.1%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 12 months ago, in November 2025, for Gemini 3 Pro. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (14 entries).
- Pass Models scored 3 models we track, plus 11 other entries on its leaderboard.
- Warning Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro.
- Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3 Pro (#56).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on github.com: 3 rows for models we track and 11 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.