ActiveSciBench-GRN (LLM-AutoSciLab) - Edge F1: leaderboard
Edge-level recovery on ActiveSciBench-GRN (Kabra et al., 2026), where the system designs perturbation experiments under a fixed budget to recover a hidden signed gene-regulatory graph, over 45 tasks. Edge F1 in percent against the true graph, crediting partial structure; higher is better, with the listed model driving the LLM-AutoSciLab framework.
Metric: Edge F1 (%; directed-edge recovery against the hidden regulatory graph over the 45 ActiveSciBench-GRN tasks; 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 | 72.49 | #1627 |
| 2 | Qwen 3 32B | 62.7 | #960 |
| 3 | Qwen 3 14B | 51.19 | #1235 |
| 4 | Qwen 3 4B | 49.56 | #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-edge-f1 · How It Works · Data refreshed daily, snapshot 2026-11-12.