Gemini 3.0 is absolutely lazy asf when it comes to analysis, reviews etc. and hallucinates wildly, anyone else having the same experience?

I gave it a full manuscript for review (about 15 pages), it missed all important issues, gave a lazy asf assessment, hallucinated figure references and glazed me unnecessarily. The same prompt and file on GPT-5.1 high (API) actually discovered the critical issues and gave a much more thorough assessment. Is Google serving some quantized version to save money? I am using aistudio and selected the high thinking variant so I think I am using the right model.