List your company

Mathematical Reasoning on OlympiadBench (MV, EM, PK)

37.86EM ScoreEntropy reweighting
0.607210.278619.9529.6214Apr 2, 2026Apr 8, 2026Apr 14, 2026Apr 20, 2026Apr 26, 2026May 2, 2026May 8, 2026

Evaluation results

MethodLinks
Entropy reweightingBackbone=Qwen3-8B-Base, Layer=Deep (Layer 31)2026.0537.86——
Explorer-only, successful seeds (diagnostic)Backbone=Qwen3-8B-Base2026.0537.18——
Entropy reweightingBackbone=Qwen3-8B-Base, Layer=Shallow (Layer 8)2026.0536.95——
Entropy reweightingBackbone=Qwen3-8B-Base, Layer=Mid (Layer 20)2026.0536.48——
Full-token DAPOBackbone=Qwen3-8B-Base2026.0534.24——
High Prediction Entropy-onlyBackbone=Qwen3-8B-Base2026.0533.67——
Random-20%Backbone=Qwen3-8B-Base2026.0532.83——
Anchor-onlyBackbone=Qwen3-8B-Base2026.0530.47——
I-PPOBackbone=Phi3 4B2026.0424.2324.2943.47
I-PPOBackbone=Qwen2.5 3B2026.0422.1123.2844.03
I-PPOBackbone=LLaMa2026.0421.5922.8743.1
I-PPOBackbone=LLaMa3 8B2026.0421.5922.8743.1
Explorer-only, all seedsBackbone=Qwen3-8B-Base2026.0520.18——
I-PPOBackbone=Gemma2026.0417.8222.5134.15
I-PPOBackbone=Gemma2 2B2026.0417.8222.5134.15
PPOBackbone=Phi3 4B2026.0416.6321.3341.35
PPOBackbone=LLaMa2026.0416.4920.9441.19
PPOBackbone=LLaMa3 8B2026.0416.4920.9441.19
GRPOBackbone=LLaMa2026.0415.718.1938.22
PPOBackbone=Qwen2.5 3B2026.0415.3220.6141.03
PPOBackbone=Gemma2026.0411.8516.2832.04
PPOBackbone=Gemma2 2B2026.0411.8516.2832.04
BaselineBackbone=Phi3 4B2026.0411.7615.3235.53
BaselineBackbone=LLaMa2026.0410.5815.7235.81
BaselineBackbone=LLaMa3 8B2026.0410.5815.7235.81
GRPOBackbone=Gemma2026.0410.2615.0430.1
BaselineBackbone=Qwen2.5 3B2026.049.7814.6436.08
I-PPOBackbone=Rho2026.047.275.8220.47
I-PPOBackbone=Rho 1B2026.047.275.8220.47
BaselineBackbone=Gemma2026.046.410.5528.52
BaselineBackbone=Gemma2 2B2026.046.410.5528.52
PPOBackbone=Rho2026.044.455.7620.21
PPOBackbone=Rho 1B2026.044.455.7620.21
GRPOBackbone=Rho2026.043.593.1215.71
BaselineBackbone=Rho2026.042.042.8212.59
BaselineBackbone=Rho 1B2026.042.042.8212.59