MusTBENCH - Mood Trajectory Reasoning: leaderboard
MusTBench (2026) tests temporal grounding of audio-language models in full-length music tracks, zero-shot. Mood trajectory reasoning: give the time intervals where the music reaches its highest or lowest emotional intensity. Mean of temporal IoU and temporal F1, in percent; higher is better.
Metric: Mood Trajectory Reasoning (MTR, 250 questions): the model gives the time intervals where the music reaches its highest or lowest emotional intensity; score is the mean of temporal IoU and temporal F1 between predicted and ground-truth intervals (%, 0-100), on the MusTBench test set (full-length MTG-Jamendo and Slakh2100 tracks, music-expert-validated questions), zero-shot with the full track and a task instruction, temperature 0, unparseable timestamps and intervals counted wrong; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 12 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 33.4 | #145 |
| 2 | Gemini 3 Flash | 30 | #175 |
| 3 | Gemini 2.5 Flash | 29 | #463 |
| 4 | Gemini 2.5 Pro | 25.5 | #276 |
| 5 | GPT-Audio 1.5 | 17 | #501 |
| 6 | GPT-Audio | 16.7 | #607 |
| 7 | Music Flamingo | 13.6 | #866 |
| 8 | Qwen2.5-Omni-3B | 10.5 | #3519 |
| 9 | Qwen 3 Omni 30B A3B | 10.4 | #847 |
| 10 | Qwen2.5-Omni-7B | 8.5 | #1884 |
| 11 | Phi-4 Multimodal Instruct | 6.9 | #4154 |
| 12 | AF-Next | 2.8 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=mustbench-mood-trajectory-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-21.