PBSBench - Cell Multiple Choice (Out-of-Domain): 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 cells from three external datasets; higher is better.

Metric: Accuracy (%, times 100) on PBSBench cell-level out-of-domain multiple-choice questions about single white-blood-cell crops from peripheral blood smears (1,493 questions on cells from three external datasets); questions generated from GPT-4o morphology captions of hematopathologist-labeled cells; greedy decoding, model chat templates; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 15 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-4o93#754
2Gemini 2.5 Pro92#280
3InternVL3.5-8B91#1618
4Claude Sonnet 4.590#267
5Qwen 2 VL 7B Instruct87#2581
6Qwen 2.5 VL 7B Instruct84#1919
7PathGen-LLaVA78#1421
8DeepSeek-VL-7B59#7442
9llava-v1.6-mistral-7B58#4027
10LLaVA-1.6-Vicuna-7B55
11Idefics3-8B-Llama344#5592
12Quilt-LLaVA39#5009
13LLaVA-Med26
14Med-Flamingo26
15BLIP2-OPT-6.7B22

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=pbsbench-cell-multiple-choice-out-of-domain · How It Works · Data refreshed daily, snapshot 2026-11-22.