Native Robot Reasoning - General Baseline - GPQA Diamond: leaderboard

Table6 base=Qwen/Qwen3.8-27B. evalscope1.13.0, vLLM BF16, thinking, temp1, top_p.95, top_k20; question/sample count 198×8. Base and trained model evaluated on identical questions; only released base admitted. Metric: Accuracy / native task score (%). Source: https://arxiv.org/html/2611.18562v1

Metric: Accuracy / native task score (%). Source: arxiv.org. 1 model tracked. Updated 17 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen3.8-27BThinking89.6

Interactive version: theaggregate.ai/benchmark?slug=native-robot-reasoning-general-baseline-gpqa-diamond · How It Works · Data refreshed daily, snapshot 2026-11-17.