84.6%
#1
ARC-AGI-2 score of Gemini 3 Deep Think (2/26), the top system in February 2026; Gemini 3.1 Pro (Preview) 77.1%
What the chart shows
ARC-AGI-2 tests whether a system can infer a rule from a few example grid puzzles and apply it to a new one. In February 2026 the February version of Gemini 3 Deep Think led with 84.6%, ahead of Gemini 3.1 Pro (Preview) at 77.1% and GPT-5.2 (Refine) at 72.9%.
The same model can score very differently by version and setting: the earlier Gemini 3 Deep Think preview scored 45.1%, and four Claude Opus 4.6 settings ranged from 64.6% to 69.2%. The top 15 entries span 46.1 points.
Key numbers
| Measure | Value |
|---|---|
| Gemini 3 Deep Think (2/26) | 84.6% |
| Gemini 3.1 Pro (Preview) | 77.1% |
| GPT-5.2 (Refine) | 72.9% |
| Claude Opus 4.6 (120K, High) | 69.2% |
| Gemini 3 Deep Think (Preview) | 45.1% |
| GPT-5.2 Pro (Medium), 15th | 38.5% |
How to read it
- Top 15 entries on the ARC Prize leaderboard (arcprize.org) as of February 2026. Entries are model-and-setting combinations, so some models appear several times.
- Text in brackets is the version, reasoning effort, context budget or refinement method, as named by the leaderboard.
- The leaderboard also reports cost per task, which this chart does not show.
Source: ARC-AGI-2 Leaderboard, 2026, via the AI Index Report 2026 (Stanford HAI), figure 2.4.3: ARC-AGI-2. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart
Discussion