List your company

Mathematical Reasoning on GPQA (mean@8)

43.81Mean@8Prompt-GDRO
25.693230.396635.139.8034Jan 27, 2026

Evaluation results

MethodLinks
Prompt-GDROBackbone=Qwen3-8B-Base2026.0143.81
Rollout-GDROBackbone=Qwen3-8B-Base2026.0143.31
GRPO (Baseline)Backbone=Qwen3-8B-Base2026.0142.25
Prompt-GDROBackbone=Qwen3-4B-Base2026.0140.88
Rollout-GDROBackbone=Qwen3-4B-Base2026.0138.51
GRPO (Baseline)Backbone=Qwen3-4B-Base2026.0135.54
Prompt-GDROBackbone=Qwen3-1.7B-Base2026.0129.29
Rollout-GDROBackbone=Qwen3-1.7B-Base2026.0127.72
GRPO (Baseline)Backbone=Qwen3-1.7B-Base2026.0126.39