X-PCR - Lesion Characterization: leaderboard

X-PCR (2026) tests multimodal models on ophthalmic diagnosis as a six-stage reasoning chain over six imaging modalities, with multiple-choice questions built from 51 public datasets and checked by ophthalmologists. Accuracy at the lesion characterization stage, in percent; higher is better.

Metric: Accuracy (%) at stage 3, characterizing lesion morphology and distribution, on X-PCR's ophthalmic multiple-choice VQA (26,415 images from 51 public datasets, six imaging modalities, 52 diseases), zero-shot, temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-581.02#178
2LLaVA-Next-LLaMA3-8B79.44#5654
3Qwen 2.5 VL 7B78.15#1862
4Qwen 3 VL 8B75.8#917
5InternVL3-32B74.17
6Lingshu-7B73.37#1635
7Qwen 3 VL 30B A3B72.89#784
8Qwen 2.5 VL 72B72.51#781
9MedGemma-27B-IT72.28#1350
10Gemini 2.5 Pro71.15#276
11InternVL3-8B69.42#1782
12GLM-4.5V65.32#754
13GPT-5 Nano65.08#906
14Gemini 2.5 Flash62.82#463
15Claude Haiku 4.562.79#583

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=x-pcr-lesion-characterization · How It Works · Data refreshed daily, snapshot 2026-11-21.