List your company

Mathematical Reasoning on OlympiadBench (Pass@1 (%))

68.1Pass@1 AccuracyOLMo-32B-Instruct (Teacher)
36.69244.8465361.154May 31, 2026Jun 4, 2026Jun 9, 2026Jun 14, 2026Jun 18, 2026Jun 23, 2026Jun 28, 2026

Evaluation results

MethodLinks
OLMo-32B-Instruct (Teacher)Model Category=Off-the-shelf2026.0668.1
MIPUModel=Qwen3-4B, Rollout Quantization=FP82026.0667.86
LR-decayModel=Qwen3-4B, Rollout Quantization=FP82026.0667.26
GRPO BaselineModel=Qwen3-4B, Rollout Quantization=FP82026.0664.89
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Gemini-2.5-flash, Method=OmniOPD2026.0564.58
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Claude-4.5-haiku, Method=OmniOPD2026.0564.35
Claude-4.5-haikuStudent Model=-, Teacher Model=Claude-4.5-haiku, Method=Base2026.0562.69
SFTStudent Model=Qwen3-4B, Teacher Model=Gemini-2.5-flash, Method=SFT2026.0562.61
Token Zones + KL AnnealingModel Category=Proposed (SEAD components)2026.0662.5
MISModel=Qwen3-4B, Rollout Quantization=FP82026.0662.5
SEAD (full)Model Category=Proposed (SEAD components)2026.0662.1
Gemini-2.5-flashStudent Model=-, Teacher Model=Gemini-2.5-flash, Method=Base2026.0561.86
OPD (RKL k=1)Model Category=Baseline2026.0660.1
Token Zones onlyModel Category=Proposed (SEAD components)2026.0660.1
MIPUModel=Qwen3-1.7B, Rollout Quantization=FP82026.0659.52
LR-decayModel=Qwen3-1.7B, Rollout Quantization=FP82026.0658.93
GRPOModel Category=Baseline2026.0658.8
OmniOPDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Method=OmniOPD2026.0558.64
KL Annealing onlyModel Category=Proposed (SEAD components)2026.0658.5
MISModel=Qwen3-1.7B, Rollout Quantization=FP82026.0658.33
OLMo-7B-Instruct (Student)Model Category=Off-the-shelf2026.0658.2
OPDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Method=OPD2026.0558.08
OPSDModel Category=Baseline2026.0657.8
SFTStudent Model=Qwen3-4B, Teacher Model=Claude-4.5-haiku, Method=SFT2026.0557.71
GRPO BaselineModel=Qwen3-1.7B, Rollout Quantization=FP82026.0656.55
Qwen3-30B-A3B-Inst.Student Model=-, Teacher Model=Qwen3-30B-A3B-Inst., Method=Base2026.0556.38
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B-Inst., Method=OmniOPD2026.0556.05
Qwen3-1.7B (GRPO)Student Model=Qwen3-1.7B, Teacher Model=-, Method=GRPO2026.0555.67
Qwen3-32BStudent Model=-, Teacher Model=Qwen3-32B, Method=Base2026.0552.93
SFTStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-32B, Method=SFT2026.0552.19
Qwen3-4B (GRPO)Student Model=Qwen3-4B, Teacher Model=-, Method=GRPO2026.0552.19
OmniOPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-32B, Method=OmniOPD2026.0550.27
OPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B-Inst., Method=OPD2026.0550.14
OPDStudent Model=Qwen3-4B, Teacher Model=Qwen3-32B, Method=OPD2026.0548.96
SFTStudent Model=Qwen3-4B, Teacher Model=Qwen3-32B, Method=SFT2026.0547.19
SFTStudent Model=Qwen3-4B, Teacher Model=Qwen3-30B-A3B-Inst., Method=SFT2026.0546.29
Qwen3-4BStudent Model=Qwen3-4B, Teacher Model=-, Method=Base2026.0540.32
Qwen3-1.7BStudent Model=Qwen3-1.7B, Teacher Model=-, Method=Base2026.0537.9