DialBGM (One-Line Summary): leaderboard
Background music selection on DialBGM (2026) when the model sees a one-sentence GPT-4o summary of the conversation instead of the full dialogue: for each of 1,200 dialogues it scores four candidate clips, and its top choice is compared with the human rank-1 clip. Hit@1 in percent (chance 25); higher is better.
Metric: Hit@1 (%, times 100): share of dialogues where the model's top-scored clip is the human rank-1 clip (chance 25), with a one-sentence GPT-4o summary of the dialogue as input, on DialBGM, 1,200 everyday multi-turn dialogues each paired with four candidate music clips ranked by human annotators for suitability as background music, the model scoring each audio clip; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 7 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 34.74 | #276 |
| 2 | Gemini 2.5 Flash | 32.28 | #463 |
| 3 | GPT-Audio | 31.92 | #607 |
| 4 | Qwen2.5-Omni | 31.53 | #1588 |
| 5 | Music Flamingo | 28.77 | #866 |
| 6 | Audio Flamingo | 27.99 | |
| 7 | Phi-4 Multimodal Instruct | 26.22 | #4154 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=dialbgm-one-line-summary · How It Works · Data refreshed daily, snapshot 2026-11-21.