I dont know how the massive context works but Caching is certainly a thing and cheaper: https://ai.google.dev/gemini-api/docs/caching?lang=python https://ai.goo

I dont know how the massive context works but Caching is certainly a thing and cheaper: https://ai.google.dev/gemini-api/docs/caching?lang=python https://ai.google.dev/gemini-api/docs/caching?lang=python Gemini is better than Sonnet if you have broad questions that concern a large codebase, the context size seems to help there. People also use subagents for specific purposes to keep each context size manageable, if possible. On a related note I think the agent metaphor is a bit harmful because it suggests state while the LLM is stateless.