PathBind - Image PTA D1: leaderboard
PathBind paper v1 full 18-model answer-accuracy cohort; PathBind-VQA contains 1,500 questions across six diagnostic dimensions and PathBind-PTA contains 600 questions across five dimensions. Legacy pathology datasets retain their exact source splits. Image and text-only prompts share the question and output format; the latter removes only the image. Fixed decoding parameters are not numerically disclosed. API models use official endpoints; open models run locally on eight RTX 4090 GPUs. Unspecified checkpoint variants retain source-qualified identities. The arrows are improvements over text-only, never the reported accuracy. Image input; PTA D1. Metric: Score (%).
Metric: Score (%). Source: arxiv.org. Saturation forecast: Around October 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 68.3 | #82 |
| 2 | GPT-5.4 | 53.3 | #84 |
| 3 | InternVL3-8B | 50.8 | #621 |
| 4 | InternVL3-38B | 50 | #414 |
No result here: #5 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=pathbind-image-pta-d1 · How It Works · Data refreshed daily, snapshot 2026-10-26.