List your company

Mathematical Reasoning on AIME 2025 (Mean@32 accuracy)

66Mean@32 AccuracyHTPO
21.17632.81344.4556.087May 8, 2026

Evaluation results

MethodLinks
HTPOBackbone=Qwen3-8B-Instruct2026.0566
DAPOBackbone=Qwen3-8B-Instruct2026.0564.7
GSPOBackbone=Qwen3-8B-Instruct2026.0563.7
SAPOBackbone=Qwen3-8B-Instruct2026.0563.6
80/20-RuleBackbone=Qwen3-8B-Instruct2026.0563.4
BAPOBackbone=Qwen3-8B-Instruct2026.0559.2
GRPO†Backbone=Qwen3-8B-Instruct2026.0558.6
HTPOBackbone=Qwen3-8B-Base2026.0530.4
GSPOBackbone=Qwen3-8B-Base2026.0530.3
SAPOBackbone=Qwen3-8B-Base2026.0528.4
80/20-RuleBackbone=Qwen3-8B-Base2026.0525.9
BAPOBackbone=Qwen3-8B-Base2026.0524.7
DAPOBackbone=Qwen3-8B-Base2026.0523.7
GRPO†Backbone=Qwen3-8B-Base2026.0522.9