ARC-AGI-3
Interactive rule discovery, planning and action efficiency on unfamiliar game environments.
How it is run
Verified semi-private ARC-AGI-3 scores only. Best published configuration per release is selected separately for standard and provider-adapter runs.
What it cannot tell you
Standard and provider adapters are separate fields; public-demo and community submissions are excluded. Scores measure this task protocol, not general intelligence.
- Leader
- GPT-6 Astra (Max)62.71% · ARC Prize · Standard results, grouped here by release
- Top-ten spread
- 62.28%Leader minus tenth place, or last place in smaller fields.
- Median
- 0.51%of the 17 rows published
- No published result
- 35of 46 models on this site
ARC Prize · Standard results by release
Verified semi-private set; best published configuration per release within this scaffold. Public demo and community submissions excluded. Standard and adapter results remain separate.
| # | Model | Score | Effort |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 62.71% | Standard |
| 2 | Claude Opus 5 (High) | 30.16% | Standard |
| 3 | Gemini 3.8 Flash (High) | 10.37% | Standard |
| 4 | GPT-5.6 Sol (Max) | 7.78% | Standard |
| 5 | GPT-6 Sol (Max) | 4.62% | Standard |
| 6 | Grok 4.6 (XHigh) | 2.11% | Standard |
| 7 | Claude Opus 4.8 (High) | 1.52% | Standard |
| 8 | GPT-5.6 Terra (Max) | 0.80% | Standard |
| 9 | Anthropic Opus 4.6 (Max) | 0.51% | Standard |
| 10 | GPT-5.5 (High) | 0.43% | Standard |
| 11 | Gemini 3.1 Pro (Preview) | 0.42% | Standard |
| 12 | Grok 4.5 (Medium) | 0.32% | Standard |
| 13 | GPT-5.4 (High) | 0.21% | Standard |
| 14 | GPT-6 Luna (Medium) | 0.19% | Standard |
| 15 | Opus 4.7 (High) | 0.18% | Standard |
| 15 | GPT-5.6 Luna (Max) | 0.18% | Standard |
| 17 | Grok 4.20 (Beta Reasoning) | 0.09% | Standard |
ARC Prize · Provider adapter results by release
Verified semi-private set; best published configuration per release within this scaffold. Public demo and community submissions excluded. Standard and adapter results remain separate.
| # | Model | Score | Effort |
|---|---|---|---|
| 1 | GPT-6 Astra - Provider Adapter (High) | 99.95% | Provider adapter |
| 2 | Gemini 3.8 Flash - Provider Adapter (High) | 35.00% | Provider adapter |
| 3 | GPT-6 Sol - Provider Adapter (Max) | 23.03% | Provider adapter |
| 4 | GPT-6 Luna - Provider Adapter (Max) | 0.59% | Provider adapter |
No published result tracked
35 of the 46 models on this site have no published number on this benchmark, so their cells in the home table stay blank rather than reading as zero.
Sources
Every page used for this benchmark, read on 17 October 2026.
- Benchmark site or the page that publishes itarcprize.org/leaderboard
- Independent resultsarcprize.org/leaderboard