Benchmarks / Reasoning
ARC-AGI-3
ARC-AGI-3 semi-private set: interactive game-like environments where the agent must learn rules on the fly; percent of possible score.
ARC-AGI-3
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-6 Astra | high | 99.9% |
| 2 | Gemini 4 | low | 97.5% |
| 3 | GPT-6.1 Sol | xhigh | 96.4% |
| 4 | Gemini 3.8 Flash | high | 35.0% |
| 5 | Opus 5 | high | 30.2% |
| 6 | GPT-6 Sol | max | 23.0% |
| 7 | Grok 4.7 | high | 10.0% |
| 8 | GPT-5.6 Sol | max | 7.8% |
| 9 | DeepSeek V4.1 Flash | low | 3.3% |
| 10 | Haiku 5.5 | max | 2.8% |
| 11 | Grok 4.6 | xhigh | 2.1% |
| 12 | Opus 4.8 | high | 1.5% |
| 13 | GPT-5.6 Terra | max | 0.8% |
| 14 | GPT-6 Luna | max | 0.6% |
| 15 | Opus 4.6 | max | 0.5% |
| 16 | GPT-5.5 | high | 0.4% |
| 17 | Gemini 3.1 Pro | high | 0.4% |
| 18 | Grok 4.5 | medium | 0.3% |
| 19 | GPT-5.4 | high | 0.2% |
| 20 | Opus 4.7 | high | 0.2% |
| 21 | GPT-5.6 Luna | max | 0.2% |
| 22 | Grok 4.20 | xhigh | 0.1% |