PBSBench - Cell Multiple Choice: leaderboard

PBSBench (2026) tests vision-language models on peripheral blood smear hematopathology, with questions about single white blood cell crops. Multiple-choice accuracy in percent on in-domain cells; higher is better.

Metric: Accuracy (%, times 100) on PBSBench cell-level in-domain multiple-choice questions about single white-blood-cell crops from peripheral blood smears (1,088 questions); questions generated from GPT-4o morphology captions of hematopathologist-labeled cells; greedy decoding, model chat templates; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1InternVL3.5-8B91#1604
2GPT-4o90#751
3Gemini 2.5 Pro88#276
4Qwen 2 VL 7B Instruct81#2573
5PathGen-LLaVA78#1406
6Qwen 2.5 VL 7B Instruct76#1902
7Claude Sonnet 4.572#261
8llava-v1.6-mistral-7B52#3999
9DeepSeek-VL-7B51#7399
10LLaVA-1.6-Vicuna-7B46
11Quilt-LLaVA36#4973
12Idefics3-8B-Llama335#5542
13LLaVA-Med29
14Med-Flamingo27
15PA-LLaVA24

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

Interactive version: theaggregate.ai/benchmark?slug=pbsbench-cell-multiple-choice · How It Works · Data refreshed daily, snapshot 2026-11-21.