Darwin R3 Card - EvasionBench: leaderboard
Darwin R3 author-run EvasionBench, exact native outcome and singleton roster. inspect-ai task from the dataset eval.yaml, temperature 1.0, top_p 0.95, 8,192 max tokens, thinking on, single run. Source pinned to immutable card revision; no score or setting borrowed from R1 or parent. Source has one model for this protocol: preserve the observation, and record the producer’s actual singleton admission instead of inventing peer values.
Metric: Native harness score (%). Source: huggingface.co. 1 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Darwin-180B-RSI-R3 (Darwin R3 Card Run: EvasionBench;settings=source-pinned) | 77.83 |
Interactive version: theaggregate.ai/benchmark?slug=darwin-r3-card-evasionbench · How It Works · Data refreshed daily, snapshot 2026-11-12.