Explore the interactive model lab
AI benchmark evidence and Pareto comparisons
These public-task snapshots are derived from licensed ARC Prize records. They are not official leaderboard scores or broad production-quality rankings. Missing-task filtering and different compute budgets limit conclusions.
ARC-AGI-1 · matched public subset
SIENG-derived mean scores on 328 of 400 public tasks (82% coverage), shared by all five configurations. Source dataset last modified 2026-06-04. Not the official semi-private leaderboard. Missing-task filtering can bias the comparison; attempts and reasoning budgets vary.
| Configured model | Mean task score | Historical USD / task |
|---|---|---|
| GPT-5.2 · highgpt-5.2-2025-12-11 · high · sample run 2025-12-08 · request settings in source | 90.24% | $0.3786 |
| Claude Opus 4.8 · highclaude-opus-4-8 · high · sample run 2026-05-29 · request settings in source | 96.65% | $0.5466 |
| Claude Opus 4.6 · high, 120K capclaude-opus-4-6 · high · sample run 2026-02-04 · request settings in source | 98.17% | $0.7798 |
| Gemini 3 Flash Preview · highgemini-3-flash-preview · high · sample run 2025-12-15 · request settings in source | 88.87% | $0.1349 |
| Gemini 3.1 Pro Previewgemini-3.1-pro-preview · publisher configuration · sample run 2026-02-09 · request settings in source | 96.95% | $0.3611 |
Pinned publisher data · ARC Prize data · MIT
ARC-AGI-2 · matched public subset
SIENG-derived mean scores on 108 of 120 public tasks (90% coverage), shared by all five configurations. Source dataset last modified 2026-06-04. Not the official semi-private leaderboard. Missing-task filtering can bias the comparison; attempts and reasoning budgets vary.
| Configured model | Mean task score | Historical USD / task |
|---|---|---|
| GPT-5.2 · highgpt-5.2-2025-12-11 · high · sample run 2025-12-08 · request settings in source | 43.52% | $1.3694 |
| Claude Opus 4.8 · highclaude-opus-4-8 · high · sample run 2026-05-29 · request settings in source | 80.40% | $2.8626 |
| Claude Opus 4.6 · high, 120K capclaude-opus-4-6 · high · sample run 2026-02-04 · request settings in source | 83.18% | $3.5074 |
| Gemini 3 Flash Preview · highgemini-3-flash-preview · high · sample run 2025-12-15 · request settings in source | 37.81% | $0.2324 |
| Gemini 3.1 Pro Previewgemini-3.1-pro-preview · publisher configuration · sample run 2026-02-09 · request settings in source | 88.43% | $0.9753 |
Pinned publisher data · ARC Prize data · MIT
Complete provenance, hashes and matched task IDs · Attribution and license