Fri 13 Nov
LiveGet it free
ZOTPAPER

News without the noise

← All concepts

vision-language model

By merging visual perception with language understanding, these models can describe images, answer questions about them, or generate text based on visual input. They process both modalities jointly, enabling tasks that require combined visual and textual reasoning.

The studies examine vision-language models as components within larger systems, often serving as planners or teachers. They are combined with reinforcement learning, constraints, or tool calls to improve performance in areas like spatial reasoning, safe actions, and document understanding. This pairing highlights that these models are most effective when their inherent limitations are addressed through additional training or structure.

Papers
251
Released code
12
First seen
Mar 2026
Latest
Nov 2026

234 papers in the last two months, against 12 in the two before.

Start here

Three papers to read first, in the order they came out
  1. 1Visual AI models often can't count animals or tell individuals apart

    Vision-language models handle coarse biological categories but fail at fine taxonomic ranks, counting, trait measurement and individual re-identification, and getting bigger mostly does not help.

    Nov 2026
  2. 2AI models can read diagrams correctly and still miscount their contents

    Gemini models usually fail to count diagrams because of computation rather than perception, but reading errors become the bottleneck on larger drawings.

    Nov 2026
  3. 3Medical AI segmentation models exploit size hints leaked from benchmark captions

    In medical image segmentation, the benefit of text captions comes largely from 'extent leakage', not language understanding.

    Nov 2026

How fast it is moving

JunJulAugSeptOctNov so far

Often studied with

benchmarkA standard set of tasks or data used to evaluate and compare the performance of AI systems.67 togetherfine-grained spatial reasoningReasoning precisely about object positions, orientations, and spatial relations across planar, depth, and temporal dimensions for robotics and navigation.36 togetherreinforcement learningA machine learning method where an agent learns to make sequences of decisions by receiving rewards or penalties for its actions.26 together

Who is working on it

NVIDIA 15Alibaba Group 10Zhejiang University 9Shanghai Jiao Tong University 8Peking University 7Ant Group 7

The papers

Most central to this idea first, not most recent.

Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.