SVCBench - O2-Gain: leaderboard
Metric: Gaussian Precision Accuracy (0-100): per query point exp(-(prediction - truth)^2 / (2 sigma^2)) with sigma 5% of the true count, averaged over the 78 O2-Gain (new individuals within a window) questions' query points; unparseable answers score 0; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seed 1.8 | 69.2 | #136 |
| 2 | Gemini 3 Flash | 55.3 | #93 |
| 3 | Qwen 3.5 35B A3B | 53.8 | #250 |
| 4 | Qwen 2.5 VL 7B Instruct | 46.2 | #643 |
| 5 | GPT-5.4 | 35.9 | #76 |
| 6 | Kimi K2.5 | 24.4 | #139 |
| 7 | Molmo2-8B | 3.8 | #522 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=svcbench-o2-gain · How It Works · Data refreshed daily, snapshot 2026-10-11.