IMCBench: leaderboard
Image-grounded medical conversation benchmark (IMCBench, Xenochristou et al., ECML PKDD 2026): the model acts as a remote dermatology triage assistant in multi-turn chats with a simulated patient (Claude Haiku 4.5) who has a synthetic health record and shares a real, biopsy-labelled skin-lesion photo. A jury of Claude Opus 4.6 and GPT-5.2, with rubrics tuned against clinician ratings, scores clinical safety, clinical accuracy and uncertainty handling from 1 to 5, and the score is their mean, higher is better; both jury models are also among the evaluated assistants.
Metric: Overall Score (1-5; mean of LLM-jury clinical safety, clinical accuracy and uncertainty handling). Source: arxiv.org. Saturation forecast: Around April 2028. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 3.61 | #100 |
| 2 | Claude Sonnet 4.6 | 3.3 | #158 |
| 3 | GPT-5.2 | 3.29 | #172 |
| 4 | Nova 2.0 Lite (Non-reasoning) | 3.12 | #1133 (Nova 2.0 Lite) |
| 5 | Claude Haiku 4.5 | 3.05 | #568 |
| 6 | Llama 4 Scout | 2.99 | #1787 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=imcbench · How It Works · Data refreshed daily, snapshot 2026-11-12.