Hx2Dx Conversational - Correct Diagnoses: leaderboard
Hx2Dx Conversational - Correct Diagnoses: Correct diagnoses out of 1000 clinical cases. Hx2Dx arXiv 2611.14122v1: 1000 clinical cases, conversational physician agents, up to 50 question/investigation turns. Patient/Examiner/Data: Gemini 3.5 Flash Lite, temperature 0. Physician temperature 0; built-in thinking enabled at provider defaults on the four reasoning models. Table 2 gives counts, not rates. GPT Luna 6 is Table 1 GPT-6 Luna. Ministral 8B is held because its release generation/quantization cannot be established from the paper.
Metric: Correct diagnoses out of 1000 clinical cases. Source: arxiv.org. Saturation forecast: Around July 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 Flash | 366 | #119 |
| 2 | DeepSeek V4.1 Flash | 327 | #58 |
| 3 | GPT-6 Luna | 276 | #93 |
| 4 | Gemini 3.5 Flash Lite | 192 | #300 |
| 5 | Llama 4 Scout | 77 | #1787 |
| 6 | Gemma 3 27B (IT) | 62 | #1306 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=hx2dx-conversational-correct-diagnoses · How It Works · Data refreshed daily, snapshot 2026-11-12.