Google Gemini: Powerful Models, Terrible Developer Experience

Google Gemini boasts leading model capabilities, including strong coding, reasoning, and multimodal abilities, plus ultra-long context windows. However, the developer experience is abysmal. The API is split across Vertex AI and Google AI Studio with inconsistent functionality; documentation is poor and outdated; the Vertex AI SDK lacks API key authentication and support for fine-tuned models; and prefix caching is incredibly unfriendly. Despite this, Gemini models offer cost advantages in long context and multimodal tasks, meaning developers may still need to use them, often relying on third-party tools like the Vercel AI SDK to mitigate the poor experience.