Since Gemini Cli doesn’t yet support some features in Claude Code, I’m using gemini 2.5 pro in CC through Claude Code Router.
Currently observed issues:
- Hallucination
I asked it to tell me where to start reading for the code of a certain feature, while referencing a folder(@frontend/). However, it only looked at the filenames in that folder without actually reading the contents of the files, and then began fabricating information. Sonnet wouldn’t do this.
(Another hallucination scenario: I was in AI Studio and tell it search for content in English, but I forgot to enable the search tool. It then told me it had searched and blah blah…). - Lack of agentic?
I was in plan mode and included some file names in the prompt(no reference, just name, so CC won’t directly put the file content into the LLM’s context). The plan it generated actually included reading those files instead of directly reading them and then creating a more detailed plan. Sonnet wouldn’t do this, Sonnet will read these files and make a deepen plan