The model scores 52.8% on CursorBench at default effort, performing on par with Opus 5.5.
Cursor developers can now use Gemini 4 Argon directly within the editor at matching performance to Opus 5.5.
Read this as text
Back to the AI news