List your company

Mathematical Reasoning on AIME 2026 (avg@16 accuracy, pass@8)

86Accuracy (avg@16)THINC-4B
10.91230.40649.969.394Apr 21, 2026Apr 23, 2026Apr 26, 2026Apr 29, 2026May 2, 2026May 5, 2026May 8, 2026

Evaluation results

MethodLinks
THINC-4BReasoning Type=THINC Student2026.0586—
Qwen3-235B-A22B-ThinkingReasoning Type=NL-Only2026.0582.1—
ASTER-4BReasoning Type=Tool-Integrated2026.0578.8—
gpt-oss-20B (medium)Reasoning Type=NL-Only2026.0576.9—
Qwen3-4B-Thinking-2507Reasoning Type=NL-Only2026.0576.7—
Qwen3.5-27B*Reasoning Type=THINC Teacher, Python Interpreter=prompted2026.0575.6—
Qwen3-4B-Thinking-2507*Reasoning Type=Tool-Integrated, Python Interpreter=prompted2026.0572.3—
rStar2-Agent-14BReasoning Type=Tool-Integrated2026.0572.3—
OpenReasoning-Nemotron-7BReasoning Type=NL-Only2026.0572.1—
DemyAgent-4BReasoning Type=Tool-Integrated2026.0565—
ReTool-32BReasoning Type=Tool-Integrated2026.0552.7—
THINC-1.7BReasoning Type=THINC Student2026.0542.9—
TEMPOBackbone=Qwen3-14B-Base2026.0438.870
Zero-RL (PPO)Backbone=Qwen3-14B-Base2026.0438.170
Qwen3-1.7BReasoning Type=NL-Only2026.0536.2—
MiMo-Zero-RL-7BBackbone=MiMo-Zero-RL-7B2026.043552.8
Qwen3-1.7B*Reasoning Type=Tool-Integrated, Python Interpreter=prompted2026.0532.1—
TTRLBackbone=Qwen3-14B-Base2026.0431.743
TEMPOBackbone=OLMO3-7B-Base2026.0430.149.4
EMPOBackbone=Qwen3-14B-Base2026.0428.349.5
Zero-RL (PPO)Backbone=OLMO3-7B-Base2026.0426.742.8
CoRT-1.5BReasoning Type=Tool-Integrated2026.0525.8—
TEMPOBackbone=Qwen3-8B-Base2026.0424.250.7
OLMO3.1-Zero-RL-7BBackbone=OLMO3.1-Zero-RL-7B2026.042442.4
EMPOBackbone=OLMO3-7B-Base2026.0423.639.7
TTRLBackbone=OLMO3-7B-Base2026.0422.839.2
Zero-RL (PPO)Backbone=Qwen3-8B-Base2026.0421.943.7
EMPOBackbone=Qwen3-8B-Base2026.0419.433.3
Oat-Zero-7BBackbone=Oat-Zero-7B2026.0416.726.3
TTRLBackbone=Qwen3-8B-Base2026.0413.825