CogVLM2
High-resolution open VLM aimed at GPT-4V-level image understanding.
About
CogVLM2 is Zhipu AI's second-generation open visual language model family, built on Llama-3-8B with 19-billion-parameter chat variants for English and Chinese. It supports 8K context and 1344x1344 image inputs, which suits dense document and UI understanding, and the authors claimed parity with or better than GPT-4V on many benchmarks at release. The catch is licensing: weights sit under a custom CogVLM2 license plus Llama 3 terms, so commercial use needs legal review. It is also heavy, needing serious VRAM to run well.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Authors claimed GPT-4V-level or better results on most of their benchmark set at release (May 2024).
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (verified tier, metrics-v1). Evidence: Released 2024-05-20; notable open VLM from Zhipu AI/THUDM; Custom CogVLM2 license plus Llama 3 terms; commercial use needs legal review; 19B size demands serious VRAM; higher latency than 7B-class VLMs; GPT-4V parity claims are from the authors' release notes. License: Custom CogVLM2 license; Llama-3-based variants additionally bound by the Llama 3 license..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 5.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · Released 2024-05-20; notable open VLM from Zhipu AI/THUDM
- · Custom CogVLM2 license plus Llama 3 terms; commercial use needs legal review
- · 19B size demands serious VRAM; higher latency than 7B-class VLMs
- · GPT-4V parity claims are from the authors' release notes