Skip to content

CogVLM

An open-source visual language model for image conversations

Visit official site
Category AI models

What CogVLM is

The model is described as a powerful open-source visual language model named CogVLM, with CogVLM-17B holding 10 billion visual parameters and 7 billion language parameters.

What you can do with it

  • Describe image details accurately
  • Answer many types of questions
  • Perform visual grounding
  • Hold a conversation about an image

Benchmark notes

  • The material states CogVLM-17B reached state-of-the-art performance on 10 classic cross-modal benchmarks including NoCaps, Flicker30k, RefCOCO, RefCOCO+, RefCOCOg, Visual7W, GQA, ScienceQA, VizWiz VQA and TDIUC
  • The material lists second place on VQAv2, OKVQA, TextVQA and COCO captioning
  • Results are described as exceeding or matching PaLI-X 55B

Pros & cons

✓ What we like

  • Open-source visual language model
  • Image description, question answering and visual grounding in one model

! What to watch out for

  • The stated results rely on the benchmark list in the material
  • CogVLM-17B is stated to place second on VQAv2, OKVQA, TextVQA and COCO captioning

FAQ

What can the model do with an image?

The feature list names accurate image description, answering various questions and visual grounding.

What size is CogVLM-17B?

The material states 10 billion visual parameters and 7 billion language parameters.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review