ARC-AGI-2
Abstract reasoning puzzles
knowledge and reasoningactive
Definition
- Version
- not in the notes
- Category
- knowledge and reasoning
- Maintainer
- ARC Prize
- Status
- active
- Home page
- https://arcprize.org
Sources
- Best AI models, October 2026: Opus 5.5, GPT-6, Gemini slash-digital.io https://slash-digital.io/en/insights/ai-models-2026/
Scores at a Glance
Score in %
Higher is better for this unit.
1
GPT-6 AstraOpenAI$1.12/task
Claude Opus 5.5Anthropichigh effort, $0.408/task
GPT-6 AstraOpenAI$1.12/task95 %
vendor-reported2Claude Opus 5.5Anthropichigh effort, $0.408/task93.3 %
vendor-reported0255075100
Where this comes from
Each bar is one model, with its most independent recorded score. Hatched bars are vendor-reported; the word beside each bar says who ran it. Settings and dates differ between rows; the Scores table below has every result.
Scores
| Model | Score | Who ran it | Setting | View or population | Date | Source |
|---|---|---|---|---|---|---|
| GPT-6 Astra | 95 % | vendor-reported | $1.12/task | – | – | slash-digital.io |
| Claude Opus 5.5 | 93.3 % | vendor-reported | high effort, $0.408/task | – | – | slash-digital.io |
 Gemini 3 Deep Think | 84.6 % | vendor-reported | Updated Deep Think; verified by the ARC Prize Foundation per vendor | Gemini 3 Deep Think upgrade post | 12 Feb 2026 | blog.google |
| Gemini 3 Deep Think | 45.1 % | vendor-reported | With code execution; ARC Prize Verified per vendor | Gemini 3 launch post | 18 Nov 2025 | blog.google |
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "ARC-AGI-2", https://aurion.semai.id/benchmarks/arc-agi-2/, records of October 2026.