AGL1K - Non-Speech Distance Error: leaderboard
AGL1K mean great-circle distance error in km, lower is better, on the half of its recordings without human speech, so location must be inferred from environmental sounds such as birdsong, waves, traffic or bells. A refusal counts as a 10,000 km error; Beihang University, 2026.
Metric: Mean distance error on clips without speech (km). Source: arxiv.org. Saturation forecast: Around February 2027. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 3727.33 | #135 |
| 2 | Gemini 2.0 Flash | 4660.31 | #702 |
| 3 | Gemini 2.5 Pro | 4693.67 | #262 |
| 4 | Gemini 2.0 Flash (Thinking) | 4767.61 | #702 (Gemini 2.0 Flash) |
| 5 | Gemini 2.0 Flash Lite | 5492.96 | #1032 |
| 6 | Gemini 2.5 Flash | 5682.74 | #452 |
| 7 | Gemini 2.5 Flash Lite | 5917.32 | #958 |
| 8 | Gemma 3n E4B (IT) | 6825.13 | #2220 |
| 9 | Phi-4 Multimodal Instruct | 6831.08 | #3914 |
| 10 | Qwen2.5-Omni-7B | 7777.2 | #1797 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=agl1k-non-speech-distance-error · How It Works · Data refreshed daily, snapshot 2026-11-12.