List your company

Science Reasoning on GPQA Diamond

71.6AUCOAAFormat-Adaptive-Answer
39.56847.88456.264.516Jan 6, 2026

Evaluation results

MethodLinks
Format-Adaptive-AnswerBackbone=Qwen3-8B2026.0171.6
Base modelBackbone=Qwen3-8B2026.0169.4
Hard-Length 16kBackbone=Qwen3-8B2026.0168.9
Adaptive-AnswerBackbone=Qwen3-8B2026.0168.8
SFTBackbone=Qwen3-8B2026.0168.7
TWYNBackbone=Qwen3-8B2026.0167.1
Hard-Length 8kBackbone=Qwen3-8B2026.0163.5
Soft-LengthBackbone=Qwen3-8B2026.0162.9
Hard-Length 8k → 4kBackbone=Qwen3-8B2026.0161
Normalized-LengthBackbone=Qwen3-8B2026.0158.3
No-ThinkingBackbone=Qwen3-8B2026.0140.8