CogAgent (source unspecified)
Zhipu AI (Z.ai)
This row preserves the original CogAgent generative-model label used by OSWorld when the evaluated checkpoint cannot be resolved between the separately released chat and VQA variants. The public evaluation code names the chat variant in its caller but supplies a conflicting CogVLM server default, without an exact run record identifying the loaded weights. Results that explicitly identify CogAgent-Chat or CogAgent-VQA are attached to their separate checkpoint rows.
Benchmark scores
| Benchmark | Score |
|---|---|
| OSWorld | 1.1 |
| OSWorld (self-reported, Screenshot + A11y tree) | 1.3 |
| OSWorld (self-reported, Set-of-Mark) | 1.0 |