Moonshot AI releases PerceptionBench: a diagnostic benchmark for visual perception in multimodal models
Original title:PerceptionBench
AI summary
Moonshot AI has released PerceptionBench, a visual perception benchmark derived from cases where models actually failed across more than 40 existing benchmarks. It covers 10 atomic perception capabilities and 3,000 verified questions. No tested model exceeded 60% accuracy, and many correct answers could not be reproduced when questions were repeated, indicating that models are largely guessing rather than truly perceiving. PerceptionBench aims to precisely diagnose the visual perception breakdown points of multimodal AI and promote faithful, consistent visual understanding.
Key Takeaways
- Moonshot AI has released PerceptionBench,a visual perception benchmark derived from cases where…
- It covers 10 atomic perception capabilities and 3,000 verified questions.
- No tested model exceeded 60% accuracy,and many correct answers could not be reproduced when…
- PerceptionBench aims to precisely diagnose the visual perception breakdown points of multimodal AI…
Quick Facts
- Published
- Updated
- Source
- Moonshot AI:Kimi Blog
- Category
- Papers & Research
- AI score
- 70
这是一个把视觉感知拆成原子能力的基准,所有模型不及格,而且猜测多于感知的发现很扎心,做多模态的团队应该拿来测一下自家模型。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Moonshot AI releases PerceptionBench: a diagnostic benchmark for visual perception in multimodal models. July 16, 2026. https://new.nat6.net/en/news/cmrnvwztt01bebixyznbje716/