List your company
ResearchFormerly WizWand

Mathematical Reasoning on OpenMathReasoning (val)

0.597Pass@1 AccuracyGoldilocks
0.002120.156560.3110.46544Feb 16, 2026
Updated 8mo ago

Evaluation results

MethodLinks
GoldilocksBackbone=Qwen3-4B, Training Protocol=Goldilocks, Teacher Model=Qwen3-1.7B2026.020.597
Qwen3-4B (GRPO)Backbone=Qwen3-4B, Training Protocol=GRPO, Teacher Model=Qwen3-1.7B2026.020.581
GoldilocksBackbone=Phi-4-4B-Instruct, Training Protocol=Goldilocks, Teacher Model=Qwen3-1.7B2026.020.41
Phi-4-4B-Instruct (GRPO)Backbone=Phi-4-4B-Instruct, Training Protocol=GRPO, Teacher Model=Qwen3-1.7B2026.020.371
GoldilocksBackbone=Qwen2.5-1.5B, Training Protocol=Goldilocks2026.020.334
Qwen2.5-1.5B (GRPO)Backbone=Qwen2.5-1.5B, Training Protocol=GRPO2026.020.306
Phi-4-4B-InstructBackbone=Phi-4-4B-Instruct, Training Protocol=Base Model (no RL), Teacher Model=Qwen3-1.7B2026.020.254
GoldilocksBackbone=Olmo2-1B, Training Protocol=Goldilocks2026.020.149
Qwen2.5-1.5BBackbone=Qwen2.5-1.5B, Training Protocol=Base Model (no RL)2026.020.1348
Qwen3-4BBackbone=Qwen3-4B, Training Protocol=Base Model (no RL), Teacher Model=Qwen3-1.7B2026.020.124
Olmo2-1B (GRPO)Backbone=Olmo2-1B, Training Protocol=GRPO2026.020.117
Olmo2-1BBackbone=Olmo2-1B, Training Protocol=Base Model (no RL)2026.020.025