List your company

Mathematical Reasoning on BeyondAIME (pass@16)

69Pass@16Reverse KL w/ Top-K
15.627229.483643.3457.1964May 7, 2026May 12, 2026May 17, 2026May 22, 2026May 27, 2026Jun 1, 2026Jun 7, 2026

Evaluation results

MethodLinks
Reverse KL w/ Top-KBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.0669
Forward KLBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.0668
TRDBackbone=Qwen3-8B (w/ thinking), Trajectory=yr2026.0668
BaseBackbone=Qwen3-8B (w/ thinking), Trajectory=–2026.0666
TRDBackbone=Qwen3-4B-Instruct-2507, Trajectory=yr2026.0664
Reverse KLBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.0664
Forward KL w/ ClipBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.0662
TRDModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yr, Teacher Model=Qwen3-8B2026.0662
Forward KL w/ ClipBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.0661
Forward KLModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.0661
Forward KL w/ ClipModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.0661
Reverse KLBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.0660
BaseBackbone=Qwen3-4B-Instruct-2507, Trajectory=–2026.0659
BaseModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=-, Teacher Model=Qwen3-8B2026.0659
Reverse KLModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.0659
Forward KLBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.0658
Reverse KL w/ Top-KBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.0657
Reverse KL w/ Top-KModel Backbone=Qwen3-4B-Instruct-2507, Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.0657
Forward KL w/ ClipModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.0647
BaseModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=-, Teacher Model=Qwen3-8B2026.0645
TRDModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yr, Teacher Model=Qwen3-8B2026.0645
Forward KLModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.0644
Reverse KL w/ Top-KModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.0644
Reverse KLModel Backbone=Qwen3-1.7B (w/ thinking), Trajectory Type (Traj.)=yo, Teacher Model=Qwen3-8B2026.0641
TRDModel=Qwen3-4B-Instruct-2507, Traj.=yr2026.0632.6
Forward KL w/ ClipModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.0632.4
BaseModel=Qwen3-4B-Instruct-2507, Traj.=–2026.0632.2
Forward KLModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.0632
Reverse KLModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.0631.6
Reverse KL w/ Top-KModel=Qwen3-4B-Instruct-2507, Traj.=yo2026.0631.2
S-trace-0.9Backbone=Qwen3-8B2026.0527.84
GRPO(λ)-0.9Backbone=Qwen3-8B2026.0527.61
Reverse KL w/ Top-KModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.0621.3
Forward KLModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.0621.2
Forward KL w/ ClipModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.0620.8
BaseModel=Qwen3-1.7B (w/ thinking), Traj.=–2026.0620.1
TRDModel=Qwen3-1.7B (w/ thinking), Traj.=yr2026.0620.1
Reverse KLModel=Qwen3-1.7B (w/ thinking), Traj.=yo2026.0619.7
GRPOBackbone=Qwen3-8B2026.0517.68