Conceptual Reasoning Index: leaderboard
Redwood Research's index of reasoning on hard-to-verify topics, such as aligning superhuman AI and multi-agent interaction. It is a weighted mean of three datasets: rating conceptual critiques against expert human ratings (60 percent), probability consistency across related claims (20 percent) and decision-theory multiple choice (20 percent). Zero is random guessing and 100 the maximum attainable, higher is better.
Metric: Chance-Corrected Score (0-100). Source: conceptualreasoning.ai. Saturation forecast: Around January 2027. 210 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 77.87 | #3 |
| 2 | Claude Sonnet 5.5 | 77.08 | #11 |
| 3 | Claude Opus 5.5 (High) | 76.07 | #3 (Claude Opus 5.5) |
| 4 | Claude Opus 5.5 (xHigh) | 75.76 | #3 (Claude Opus 5.5) |
| 5 | Claude Sonnet 5.5 (xHigh) | 75.59 | #11 (Claude Sonnet 5.5) |
| 6 | Claude Opus 5.5 (Medium) | 75.38 | #3 (Claude Opus 5.5) |
| 7 | Claude Fable 5.1 (Medium) | 74.95 | #9 (Claude Fable 5.1) |
| 8 | Claude Fable 5.1 | 74.67 | #9 |
| 9 | Claude Opus 5 (xHigh) | 74.65 | #12 (Claude Opus 5) |
| 10 | Claude Opus 5.5 (Low) | 74.43 | #3 (Claude Opus 5.5) |
| 11 | Claude Fable 5.1 (xHigh) | 74.38 | #9 (Claude Fable 5.1) |
| 12 | Claude Opus 5 (High) | 73.94 | #12 (Claude Opus 5) |
| 13 | Claude Sonnet 5.5 (Medium) | 73.9 | #11 (Claude Sonnet 5.5) |
| 14 | GPT-6 Astra (xHigh) | 73.89 | #6 (GPT-6 Astra) |
| 15 | Claude Fable 5.1 (High) | 73.88 | #9 (Claude Fable 5.1) |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=conceptual-reasoning-index · How It Works · Data refreshed daily, snapshot 2026-10-20.