Gemini 2.5 Pro admitting own lack of capabilities

Trying to stress-test LLM Agents on RetroArch codebase (quite complex, with a lot of conditional compilation) Gemini 2.5 Pro admitted lack of own capabilities. That's actually nice behavior that I'd expect to see more from other models that don't stop and make more and more mess.