Benchmarks · reasoning
ARC-AGI-2
ARC-AGI-2 semi-private evaluation set: harder compositional grid puzzles designed to resist memorization; percent solved with verified cost per task.
87 models · 231 results · data as of 2026-10-29 · arcprize.org ↗
Leaderboard
GPT-6 Astra max
All results
231 rows| # | Model | Effort | ARC-AGI-2 | Subscription $ / task | API $ / task |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | max | 95.0% | $0.096 | $1.12 |
| 2 | GPT-6.1 Sol | max | 94.2% | $0.030 | $0.25 |
| 3 | Opus 5.5 | high | 93.3% | $0.006 | $0.41 |
| 4 | GPT-6 Astra | xhigh | 93.3% | $0.071 | $0.83 |
| 5 | Opus 5.5 | xhigh | 92.5% | $0.010 | $0.67 |
| 6 | GPT-5.6 Sol | max | 92.5% | $0.080 | $1.44 |
| 7 | GPT-6 Astra | high | 92.1% | $0.057 | $0.67 |
| 8 | GPT-6 Astra | medium | 92.1% | $0.041 | $0.48 |
| 9 | Opus 5.5 | max | 91.7% | $0.028 | $1.85 |
| 10 | GPT-6.1 Sol | high | 91.7% | $0.016 | $0.13 |
| 11 | GPT-6.1 Sol | xhigh | 91.7% | $0.021 | $0.18 |
| 12 | Gemini 4 Argon | high | 91.3% | $0.12 | $0.32 |
| 13 | Opus 5 | max | 90.4% | $0.036 | $2.06 |
| 14 | Fable 5.1 | max | 90.0% | $0.70 | $4.49 |
| 15 | Fable 5.1 | xhigh | 90.0% | $0.49 | $3.12 |
| 16 | GPT-5.6 Sol | xhigh | 90.0% | $0.057 | $1.04 |
| 17 | GPT-6 Sol | max | 89.6% | $0.027 | $0.44 |
| 18 | Fable 5 | max | 89.2% | $0.61 | $5.45 |
| 19 | Gemini 3.8 Flash | high | 89.2% | $0.016 | $0.40 |
| 20 | Fable 5.1 | high | 88.8% | $0.26 | $1.67 |
| 21 | Fable 5 | xhigh | 88.3% | $0.34 | $3.00 |
| 22 | Opus 5 | high | 88.3% | $0.025 | $1.45 |
| 23 | Fable 5 | high | 87.5% | $0.27 | $2.44 |
| 24 | Opus 5.5 | medium | 87.5% | $0.005 | $0.34 |
| 25 | Gemini 4 Argon | medium | 87.1% | $0.049 | $0.14 |
| 26 | GPT-6.1 Sol | medium | 86.7% | $0.012 | $0.10 |
| 27 | Fable 5.1 | medium | 86.3% | $0.19 | $1.22 |
| 28 | GPT-5.6 Sol | high | 85.4% | $0.041 | $0.74 |
| 29 | GPT-6 Astra | low | 85.4% | $0.036 | $0.42 |
| 30 | GPT-5.5 | xhigh | 85.0% | $0.10 | $1.87 |