AutoMedBench - Agentic Score (Lite): leaderboard
The workflow score of AutoMedBench (2026), an end-to-end medical-AI research benchmark for LLM agents, over all 24 tasks in the Lite tier, where the task brief supplies method, environment and skill scaffolding. Plan, setup, validate, inference and submit stages are weighted 25, 15, 35, 15 and 10, the first three judged by an LLM from saved artifacts and traces, on a 0 to 100 scale. Higher is better.
Metric: Agentic workflow score (0-100) over all 24 tasks in the Lite tier (method, environment and skill scaffolding provided in the task brief): Plan, Setup, Validate, Inference and Submit stage scores weighted 25/15/35/15/10, the first three judged by an LLM from saved artifacts and traces; each model is the single agent of the shared harness with default decoding; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 81.1 | #102 |
| 2 | GLM-5 | 77.9 | #260 |
| 3 | Gemini 3.1 Pro (Preview) | 71.7 | #94 |
| 4 | Qwen 3.5 397B A17B | 66.6 | #246 |
| 5 | MiniMax-M2.5 | 66.4 | #582 |
| 6 | GPT-5.4 | 66 | #123 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=automedbench-agentic-score-lite · How It Works · Data refreshed daily, snapshot 2026-11-13.