List your company

Mathematical Reasoning on MATH 500 (ArithACC, Length)

91.6Arithmetic AccuracyCoLD Base PRM
17.96837.08456.275.316Jul 21, 2025Sep 16, 2025Nov 12, 2025Jan 8, 2026Mar 6, 2026May 2, 2026Jun 29, 2026

Evaluation results

MethodLinks
CoLD Base PRMPolicy Model=Deepseek-v3.12025.0791.6646.2
CoLD Base PRMPolicy Model=Qwen3-Next-80B-A3B-Instruct2025.0790.4754.5
Vanilla Base PRMPolicy Model=Deepseek-v3.12025.0789.6953.9
Vanilla Base PRMPolicy Model=Qwen3-Next-80B-A3B-Instruct2025.0785.21,050.6
CoT-SFTBackbone=Qwen3-VL-8B2026.0669.4310.6
CoT-SFTBackbone=Qwen3-VL-4B2026.0655.8291.5
DLR (Full)Backbone=Qwen3-VL-8B2026.065429.5
DLR (Full)Backbone=Qwen3-VL-4B2026.0644.628.4
DLR (SFT)Backbone=Qwen3-VL-8B2026.0639.827.2
RoTBackbone=Qwen3-VL-8B2026.0637.864
DLR (SFT)Backbone=Qwen3-VL-4B2026.0635.427.9
DLR (Full)Backbone=Qwen3-VL-2B2026.063428.9
Direct-SFTBackbone=Qwen3-VL-8B2026.0633.50
RoTBackbone=Qwen3-VL-4B2026.0633.264
Direct-SFTBackbone=Qwen3-VL-4B2026.0629.40
DLR (SFT)Backbone=Qwen3-VL-2B2026.0629.426.7
CoT-SFTBackbone=Qwen3-VL-2B2026.0629.2324.5
RoTBackbone=Qwen3-VL-2B2026.062464
Direct-SFTBackbone=Qwen3-VL-2B2026.0620.80