List your company

Question Answering on GPQA Diamond (Accuracy and Tokens)

51.5AccuracySFT + RL
38.91642.18345.4548.717Apr 24, 2026

Evaluation results

MethodLinks
SFT + RLBackbone=Qwen3-8B2026.0451.51,382
Abstract-CoT (Warm-up + RL)Backbone=Qwen3-8B2026.0450.5174
SFT (CoT)Backbone=Qwen3-8B2026.0445.51,106
BaselineBackbone=Qwen3-8B2026.0444.9767
Abstract-CoT (Warm-up)Backbone=Qwen3-8B2026.0444.4168
Abstract-CoT (RL-only)Backbone=Qwen3-8B2026.0441.9142
SFT (no CoT)Backbone=Qwen3-8B2026.0439.4499