Gemini 4 Pro: Leaked Highlights
A 1.5M Context Window that handles huge amounts of context without pain.
In Coding and Multimodal, it has already surpassed the current strongest Claude Fable 5.1 and GPT-6 Astra in coding tests.
The new backend supports Persistent cross-session memory, and can even work without connecting to the internet through a separate API. You get continuity across conversations without writing a pile of state management yourself.
The API pricing that has leaked so far is Input $2.25 / 1M tokens, Output $11.25 / 1M tokens.