Theorem Proving on FormalML Hard Level-3
95Solved RateDreamProver (GPT-5.3-Codex)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DreamProver (GPT-5.3-Codex)Backbone LLM=GPT-5.3-Codex, Lemma Learning System=True, Pass@k=pass@322026.04 | 95 | 77.9 | |
| DreamProver (Gemini 3.1 Pro)Backbone LLM=Gemini 3.1 Pro, Lemma Learning System=True, Pass@k=pass@322026.04 | 93 | 76.2 | |
| DreamProver (Gemini 2.5 Pro)Backbone LLM=Gemini 2.5 Pro, Lemma Learning System=True, Pass@k=pass@322026.04 | 85 | 69.7 | |
| GPT-5.3-CodexBackbone LLM=GPT-5.3-Codex, Lemma Learning System=False, Pass@k=pass@322026.04 | 59 | 48.4 | |
| Gemini 3.1 ProBackbone LLM=Gemini 3.1 Pro, Lemma Learning System=False, Pass@k=pass@322026.04 | 33 | 27 | |
| Gemini 2.5 ProBackbone LLM=Gemini 2.5 Pro, Lemma Learning System=False, Pass@k=pass@322026.04 | 25 | 20.5 |