BAGEL - Xeno-canto: leaderboard

BAGEL questions on species vocalizations derived from Xeno-canto recordings, answered from text alone: frequency range, call duration, harmonic structure and modulation. Four-option closed-book questions written by GPT-4o-mini from source material that is withheld at test time; accuracy in percent, higher is better.

Metric: Accuracy (%, times 100) on the 4,242 Xeno-canto questions (text-only bioacoustic knowledge of species vocalizations) of BAGEL, closed-book four-option single-answer multiple choice (the source passage is withheld), one unified prompt, greedy decoding with seed 0, a single run per model; questions written by GPT-4o-mini from Wikipedia species articles, GloBI interaction records, bioRxiv preprints and Xeno-canto recordings, options shuffled; higher is better. Source: arxiv.org. Saturation forecast: Around May 2028. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-5.459.26#135
2Claude Opus 4.656.01#113
3Llama 3.1 8B Instruct50.45#4600
4Gemma 3 27B (IT)44.81#1324
5Qwen 3 14BNon-reasoning41.25#1269
6Qwen 3 8BNon-reasoning39.32#1873
7Phi-438.31#1931
8Qwen 3 32BNon-reasoning37.51#972
9Qwen 3 4BNon-reasoning35.81#2348
10Mistral 7B Instruct (v0.3)29.49#6750
11gemma-7B27.91#7121
12SmolLM2-360M25.04#9570
13Qwen 3 0.6BNon-reasoning24.14#8350

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=bagel-xeno-canto · How It Works · Data refreshed daily, snapshot 2026-11-21.