InterveneBench - Explanation: leaderboard
Metric: Rubric component score (0-1, times 100) for agreement of the predicted effect direction and mechanism with the social-science consensus, on InterveneBench's 74-study 2025 test set (each instance an empirical social-science study of a real policy intervention; the model proposes the causal study design from the policy context, without causal graphs); a Gemini-3-Pro grader compares the design with the expert-verified ground truth; direct inference; higher is better. Source: arxiv.org. Saturation forecast: Around August 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 | 51.7 | #131 |
| 2 | Gemini 2.5 Pro | 42.7 | #145 |
| 3 | Gemini 3 Flash | 39.3 | #93 |
| 4 | GPT-OSS-120B | 37.7 | #330 |
| 5 | DeepSeek V3.2 | 35.3 | #198 |
| 6 | Grok 4 | 34 | #169 |
| 7 | Claude 3.7 Sonnet | 34 | #241 |
| 8 | Kimi K2 | 34 | #236 |
| 9 | Qwen 3 235B A22B | 31.7 | #304 |
| 10 | Claude Sonnet 4 | 30.7 | #194 |
| 11 | GLM-4.6 | 30.3 | #246 |
| 12 | GPT-4.1 | 24.3 | #240 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=intervenebench-explanation · How It Works · Data refreshed daily, snapshot 2026-10-11.