Mathematical Reasoning on AIME 2026 (avg@16 accuracy, pass@8)
86Accuracy (avg@16)THINC-4B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| THINC-4BReasoning Type=THINC Student2026.05 | 86 | — | |
| Qwen3-235B-A22B-ThinkingReasoning Type=NL-Only2026.05 | 82.1 | — | |
| ASTER-4BReasoning Type=Tool-Integrated2026.05 | 78.8 | — | |
| gpt-oss-20B (medium)Reasoning Type=NL-Only2026.05 | 76.9 | — | |
| Qwen3-4B-Thinking-2507Reasoning Type=NL-Only2026.05 | 76.7 | — | |
| Qwen3.5-27B*Reasoning Type=THINC Teacher, Python Interpreter=prompted2026.05 | 75.6 | — | |
| Qwen3-4B-Thinking-2507*Reasoning Type=Tool-Integrated, Python Interpreter=prompted2026.05 | 72.3 | — | |
| rStar2-Agent-14BReasoning Type=Tool-Integrated2026.05 | 72.3 | — | |
| OpenReasoning-Nemotron-7BReasoning Type=NL-Only2026.05 | 72.1 | — | |
| DemyAgent-4BReasoning Type=Tool-Integrated2026.05 | 65 | — | |
| ReTool-32BReasoning Type=Tool-Integrated2026.05 | 52.7 | — | |
| THINC-1.7BReasoning Type=THINC Student2026.05 | 42.9 | — | |
| TEMPOBackbone=Qwen3-14B-Base2026.04 | 38.8 | 70 | |
| Zero-RL (PPO)Backbone=Qwen3-14B-Base2026.04 | 38.1 | 70 | |
| Qwen3-1.7BReasoning Type=NL-Only2026.05 | 36.2 | — | |
| MiMo-Zero-RL-7BBackbone=MiMo-Zero-RL-7B2026.04 | 35 | 52.8 | |
| Qwen3-1.7B*Reasoning Type=Tool-Integrated, Python Interpreter=prompted2026.05 | 32.1 | — | |
| TTRLBackbone=Qwen3-14B-Base2026.04 | 31.7 | 43 | |
| TEMPOBackbone=OLMO3-7B-Base2026.04 | 30.1 | 49.4 | |
| EMPOBackbone=Qwen3-14B-Base2026.04 | 28.3 | 49.5 | |
| Zero-RL (PPO)Backbone=OLMO3-7B-Base2026.04 | 26.7 | 42.8 | |
| CoRT-1.5BReasoning Type=Tool-Integrated2026.05 | 25.8 | — | |
| TEMPOBackbone=Qwen3-8B-Base2026.04 | 24.2 | 50.7 | |
| OLMO3.1-Zero-RL-7BBackbone=OLMO3.1-Zero-RL-7B2026.04 | 24 | 42.4 | |
| EMPOBackbone=OLMO3-7B-Base2026.04 | 23.6 | 39.7 | |
| TTRLBackbone=OLMO3-7B-Base2026.04 | 22.8 | 39.2 | |
| Zero-RL (PPO)Backbone=Qwen3-8B-Base2026.04 | 21.9 | 43.7 | |
| EMPOBackbone=Qwen3-8B-Base2026.04 | 19.4 | 33.3 | |
| Oat-Zero-7BBackbone=Oat-Zero-7B2026.04 | 16.7 | 26.3 | |
| TTRLBackbone=Qwen3-8B-Base2026.04 | 13.8 | 25 |