Theorem Proving on DeepTheorem
54False RateDeepSeek-V3.2-Thinking (Agentic)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DeepSeek-V3.2-Thinking (Agentic)2026.01 | 54 | 18 | 18 | |
| GPT-5.2 (Agentic)2026.01 | 58 | 16 | 14 | |
| DeepSeek-V3.2 (Agentic)2026.01 | 62 | 22 | 22 | |
| Gemini-3-FlashThinking level=low2026.01 | 72 | 32 | 26 | |
| Gemini-3-ProThinking level=low2026.01 | 76 | 42 | 36 | |
| Qwen-Max (Agentic)2026.01 | 76 | 22 | 22 | |
| Baseline2026.01 | 76 | 20 | 16 | |
| Claude-Sonnet-4.5 (Agentic)2026.01 | 80 | 36 | 32 |