Tag
The latest Gemini 3.5 checkpoint is criticized for poor prompt adherence, with the model ignoring instructions, using web when told not to, and overbuilding UI, raising concerns about agent reliability despite speed and intelligence.