MedCUA-Bench - Step Goals: leaderboard
Strict success rate on the step-goal half of MedCUA-Bench (2026), where each clinical GUI task spells out the sequence of clicks, fields and confirmations, isolating interface execution from clinical planning. Percent; higher is better.
Metric: Strict success rate (%) on the 216 step-goal tasks (the same tasks with the explicit sequence of clicks, form fields and confirmations spelled out) of 432 screenshot-only clinical computer-use tasks (18 scenarios in 10 medical domains: 15 synthetic reconstructions of clinical software, the real OpenEMR EHR and the OHIF imaging viewer; 216 base tasks each under an intent goal and a step goal), one run per task in the shared AgentLab harness with a 30-step budget; strict success requires task completion with no safety violation; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 23 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 56.5 | #123 |
| 2 | Claude Opus 4.7 | 54.2 | #79 |
| 3 | Claude Sonnet 4.6 | 47.7 | #161 |
| 4 | Kimi K2.6 | 33.3 | #178 |
| 5 | GPT-5 Mini | 14.8 | #348 |
| 6 | Qwen 2.5 VL 32B Instruct | 13.9 | #1053 |
| 7 | Gemini 3.1 Pro (Preview) | 4.2 | #94 |
| 8 | Gemma 3 27B | 2.8 | #1631 |
| 9 | Qwen 3.5 27B | 2.3 | #359 |
| 10 | Gemma 4 E4B | 1.4 | #2214 |
| 11 | Qwen 3.5 9B | 0.9 | #803 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=medcua-bench-step-goals · How It Works · Data refreshed daily, snapshot 2026-11-13.