ActiveSciBench-GRN (LLM-AutoSciLab): leaderboard
Active causal-graph discovery from ActiveSciBench (Kabra et al., 2026): the system chooses gene knock-up, knock-down and perturbation experiments on a simulated regulatory network under a fixed budget and recovers the hidden signed graph, over 45 tasks built from five regulatory motifs. Exact graph accuracy in percent; higher is better. The listed model drives the LLM-AutoSciLab closed-loop framework.
Metric: Exact graph accuracy (%; share of the 45 gene-regulatory-network tasks whose hidden signed causal graph is recovered exactly; budget-limited perturbation experiments driven by the LLM-AutoSciLab framework with this backbone). Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o Mini | 31.11 | #1627 |
| 2 | Qwen 3 32B | 25.88 | #960 |
| 3 | Qwen 3 14B | 22.34 | #1235 |
| 4 | Qwen 3 4B | 21.78 | #2297 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=activescibench-grn-llm-autoscilab · How It Works · Data refreshed daily, snapshot 2026-11-12.