GPAgentBench-2K - Treatment Score: leaderboard
GPAgentBench-2K (Chen et al., 2026) runs LLMs zero-shot as general practitioners in interactive consultations with a simulated patient, built from 2,428 physician-validated primary-care cases and six clinical actions. Treatment score in percent from a GPT-5.4 judge against the ground-truth treatment; higher is better.
Metric: Treatment score (%; prescribed treatment scored by a GPT-5.4 judge against the ground-truth treatment; zero-shot GP agent with six clinical actions over 2,428 physician-validated primary-care cases, GPT-4o-mini patient simulator, at most 10 turns). Source: arxiv.org. Saturation forecast: Around August 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 53.2 | #121 |
| 2 | Kimi K2.6 | 49.2 | #173 |
| 3 | Claude Opus 4.7 | 46 | #75 |
| 4 | Gemini 3.1 Pro (Preview) | 41.6 | #91 |
| 5 | DeepSeek V4 Pro | 41.2 | #181 |
| 6 | Qwen 3.7 Max | 41.1 | #142 |
| 7 | DeepSeek V4 Flash | 40.6 | #254 |
| 8 | MiMo-V2.5-Pro | 38.3 | #274 |
| 9 | Claude Sonnet 4.6 | 36 | #158 |
| 10 | GLM-5.1 | 34.3 | #175 |
| 11 | MiniMax-M2.7 | 31.3 | #519 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=gpagentbench-2k-treatment-score · How It Works · Data refreshed daily, snapshot 2026-11-12.