Epoch AI - Lmca: leaderboard
Can a model judge an argument in a field where the bottom line is still contested? 560 position texts and 1,461 arguments against them in decision theory, philosophy and AI safety, each rated by domain experts. The score is the correlation between the model’s overall ratings and the experts’, times 100 — and inter-rater agreement puts the ceiling near 85, not 100.
Metric: Score. Source: epoch.ai. 197 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | claude-sonnet-5-5_max | 68.26 | |
| 2 | claude-opus-5-5_max | 68.23 | |
| 3 | claude-sonnet-5-5_xhigh | 66.32 | |
| 4 | claude-opus-5-5_xhigh | 66.18 | |
| 5 | claude-sonnet-5-5_high | 65.93 | |
| 6 | claude-opus-5-5_high | 65.91 | |
| 7 | claude-opus-5-5_medium | 65.66 | |
| 8 | claude-fable-5-1_unknown | 65.46 | |
| 9 | claude-sonnet-5-5_medium | 65.45 | |
| 10 | claude-fable-5-1_medium | 65.32 |
Interactive version: theaggregate.ai/benchmark?slug=epoch-ai-lmca · How It Works · Data refreshed daily, snapshot 2026-10-20.