List your company

Mathematical Reasoning on AMO-Bench (average@16)

14.8Average@16GRPO
1.5925.0218.4511.879Jan 26, 2026Feb 17, 2026Mar 11, 2026Apr 2, 2026Apr 24, 2026May 16, 2026Jun 7, 2026

Evaluation results

MethodLinks
GRPOModel Scale=8B, Backbone=Qwen3, Temperature=1.2, Generation Length=38k2026.0114.8
OPSDModel Scale=8B, Backbone=Qwen3, Temperature=1.2, Generation Length=38k2026.0114.3
OPSDModel Scale=4B, Backbone=Qwen3, Temperature=1.2, Generation Length=38k2026.0113.5
Qwen3 (Base Instruct)Model Scale=8B, Temperature=1.2, Generation Length=38k2026.0113.4
SFTModel Scale=8B, Backbone=Qwen3, Temperature=1.2, Generation Length=38k2026.0112.9
GRPOModel Scale=4B, Backbone=Qwen3, Temperature=1.2, Generation Length=38k2026.0112.8
SFTModel Scale=4B, Backbone=Qwen3, Temperature=1.2, Generation Length=38k2026.0112.5
Qwen3 (Base Instruct)Model Scale=4B, Temperature=1.2, Generation Length=38k2026.0112.4
TRDModel=Qwen3-4B-Instruct-2507, Traj.=yr2026.0610.3
BaseModel=Qwen3-4B-Instruct-2507, Traj.=–2026.0610.1
Forward KL w/ ClipModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.0610.1
Forward KLModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.069.9
Reverse KL w/ Top-KModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.069.8
Reverse KLModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.068.2
OPSDModel Scale=1.7B, Backbone=Qwen3, Temperature=1.2, Generation Length=38k2026.015
GRPOModel Scale=1.7B, Backbone=Qwen3, Temperature=1.2, Generation Length=38k2026.014.5
Qwen3 (Base Instruct)Model Scale=1.7B, Temperature=1.2, Generation Length=38k2026.014.3
SFTModel Scale=1.7B, Backbone=Qwen3, Temperature=1.2, Generation Length=38k2026.013.9
Reverse KL w/ Top-KModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.063
TRDModel=Qwen3-1.7B (w/ thinking), Traj.=yr2026.063
Forward KLModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.062.4
Reverse KLModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.062.4
BaseModel=Qwen3-1.7B (w/ thinking), Traj.=–2026.062.1
Forward KL w/ ClipModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.062.1