Darwin R3 Card - GPQA Diamond - Majority of 8: leaderboard
Darwin R3 author comparison, GPQA Diamond, 198 questions, 8 samples per question, 32768 thinking tokens, temperature1.0. Single sample / mean-of-8 / majority-of-8 remain separate. R3 is trained from R1 on separate SuperGPQA practice problems; source says GPQA Diamond and the held-out set were never used in training or selection. R0 is Qwen3.8-Flash-Next; R1 and R3 are distinct merged trained checkpoints, not effort modes. Exact source table and immutable card revision pinned. No inference from the card’s mixed-protocol external leaderboard comparison.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3.8-Flash-Next (Darwin R3 Card Run: task=GPQA Diamond;metric=Majority of 8;draws=8;thinking_budget=32768;temperature=1.0) | 90.91 | |
| 2 | Darwin-180B-RSI-R3 (Darwin R3 Card Run: task=GPQA Diamond;metric=Majority of 8;draws=8;thinking_budget=32768;temperature=1.0) | 90.4 | |
| 3 | Darwin-180B-RSI (Darwin R3 Card Run: task=GPQA Diamond;metric=Majority of 8;draws=8;thinking_budget=32768;temperature=1.0) | 89.9 |
Interactive version: theaggregate.ai/benchmark?slug=darwin-r3-card-gpqa-diamond-majority-of-8 · How It Works · Data refreshed daily, snapshot 2026-11-12.