VoiceGiraffe (Cascaded Captions) - Single-Hop Perception: leaderboard
VoiceGiraffe (2026) tests hour-scale audio understanding with 1,500 four-option questions on long English and Chinese recordings. Under the cascaded-caption protocol the model captions the audio in 30-second windows and then answers from its own concatenated captions; this board is accuracy in percent on the 1,000 single-hop questions on temporal localization, semantic content, acoustic events and paralinguistics, higher is better.
Metric: Accuracy (%) on the 1,000 single-hop questions (temporal localization, semantic content, acoustic events, paralinguistics) of VoiceGiraffe on 123 English and Chinese recordings of 30 minutes to 2 hours; cascaded caption aggregation: the model captions the recording in 30-second sliding windows (speech, speaker traits, sound events, music) and then answers from the concatenated captions and the question itself, without an external reasoning model; answers are matched by regular expression; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 12 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 74 | #99 | |
| 2 | Gemini 2.5 Pro | 72.8 | #276 | |
| 3 | Qwen 3.5 Omni Plus | 66 | #254 | |
| 4 | Qwen3.5 Omni Flash | 58.4 | #602 | |
| 5 | Qwen3 Omni 30B A3B Instruct | 52.7 | #930 | |
| 6 | MOSS-Audio-8B | Thinking | 44.8 | #2150 |
| 7 | Qwen2.5-Omni-7B | 43.1 | #1884 | |
| 8 | Qwen 3 Omni 30B A3B (Thinking) | 41.5 | ||
| 9 | MiMo-Audio-7B | 41.3 | ||
| 10 | Audio Flamingo 3 | 40.5 | ||
| 11 | MOSS-Audio-8B-Instruct | 39.7 | #2017 | |
| 12 | Phi-4 Multimodal Instruct | 31.7 | #4154 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=voicegiraffe-cascaded-captions-single-hop-perception · How It Works · Data refreshed daily, snapshot 2026-11-21.