List your company

Mathematical Reasoning on AIME25 (maj@16, pass@16)

86.7Pass@16Forward KL w/ Clip
-0.004822.505145.01567.5249May 28, 2026May 29, 2026May 31, 2026Jun 2, 2026Jun 3, 2026Jun 5, 2026Jun 7, 2026

Evaluation results

MethodLinks
Forward KL w/ ClipBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.0686.7—
Reverse KLBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.0686.7—
TRDBackbone=Qwen3-8B (w/ thinking), Trajectory=yr2026.0686.7—
Reverse KL w/ Top-KBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.0686.6—
BaseBackbone=Qwen3-8B (w/ thinking), Trajectory=–2026.0683.3—
Forward KLBackbone=Qwen3-8B (w/ thinking), Trajectory=yo2026.0683.3—
PPCBackbone=Qwen3-4B2026.058063.33
TRDBackbone=Qwen3-4B-Instruct-2507, Trajectory=yr2026.0680—
BaseBackbone=Qwen3-4B-Instruct-2507, Trajectory=–2026.0676.7—
Forward KLBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.0676.7—
Forward KL w/ ClipBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.0676.7—
Reverse KLBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.0676.7—
Reverse KL w/ Top-KBackbone=Qwen3-4B-Instruct-2507, Trajectory=yo2026.0676.7—
Prompt-OnlyBackbone=Qwen3-4B2026.0576.6760
GRPOBackbone=Qwen3-4B2026.0576.6760
PTA-GRPOBackbone=Qwen3-4B2026.0576.6760
Plan-TuningBackbone=Qwen3-4B2026.0576.6756.67
BaseBackbone=Qwen3-4B2026.0573.3360
PPCBackbone=Qwen2.5-7B2026.0536.6720
PPCBackbone=Qwen2.5-Math-7B2026.0536.6723.33
GRPOBackbone=Qwen2.5-Math-7B2026.0533.3320
BaseBackbone=Qwen2.5-7B2026.053010
Prompt-OnlyBackbone=Qwen2.5-7B2026.053016.67
GRPOBackbone=Qwen2.5-7B2026.053016.67
BaseBackbone=Qwen2.5-Math-7B2026.053020
Prompt-OnlyBackbone=Qwen2.5-Math-7B2026.053016.67
Plan-TuningBackbone=Qwen2.5-7B2026.0526.6720
Plan-TuningBackbone=Qwen2.5-Math-7B2026.0526.6716.67
PTA-GRPOBackbone=Qwen2.5-7B2026.0523.3316.67
PTA-GRPOBackbone=Qwen2.5-Math-7B2026.0523.3316.67
PPCBackbone=Llama3.1-8B2026.05206.67
Plan-TuningBackbone=Llama3.1-8B2026.0513.330
GRPOBackbone=Llama3.1-8B2026.05100
PTA-GRPOBackbone=Llama3.1-8B2026.05106.67
BaseBackbone=Llama3.1-8B2026.053.330
Prompt-OnlyBackbone=Llama3.1-8B2026.053.330