Darwin R3 Card - ExtractBench: leaderboard
Darwin R3 author-run ExtractBench, exact native outcome and singleton roster. official harness, 32,768 max tokens (same as the parent), temperature 0, thinking off, 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: ExtractBench;settings=source-pinned) | 90.29 |
Interactive version: theaggregate.ai/benchmark?slug=darwin-r3-card-extractbench · How It Works · Data refreshed daily, snapshot 2026-11-12.