CausaLab (6-Node SCMs): leaderboard
Interactive causal discovery in CausaLab (Yang et al., 2026): from 2 prior measurement records and 20 interventions on a manipulator crystal governed by a hidden 6-node linear structural causal model, an LLM agent must predict the frequency of a held-out reactor crystal that shares the mechanism. Accuracy in percent over up to 50 random graph topologies, one run each; higher is better.
Metric: Reactor frequency accuracy (%; exact prediction of the held-out crystal's frequency, up to 50 hidden 6-node linear SCMs, 2 observations and 20 interventions, one run). Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 (High) | 80 | #172 (GPT-5.2) |
| 2 | GPT-5 Mini | 24 | #349 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=causalab-6-node-scms · How It Works · Data refreshed daily, snapshot 2026-11-12.