Abstract and compositional reasoning on ARC-AGI 2 (test)
51Accuracy (ARC-AGI 2 Test)EvoX
Evaluation Results
| Method | Links | |
|---|---|---|
| EvoXBackbone=Gemini-3-Pro, Inference budget=30 LLM iterations2026.02 | 51 | |
| EvoXBackbone=GPT-5, Inference budget=30 LLM iterations2026.02 | 48 | |
| OpenEvolveBackbone=Gemini-3-Pro, Inference budget=30 LLM iterations2026.02 | 45 | |
| OpenEvolveBackbone=GPT-5, Inference budget=30 LLM iterations2026.02 | 41 |