Gemini 2.5 Pro is Google’s latest sparse Mixture-of-Experts transformer that natively handles text, images, audio, and video in contexts exceeding one million tokens. It leads public coding, reasoning, and multimodal benchmarks (e.g., LiveCodeBench 69 %, GPQA 86 %) while delivering 143 tokens/s with sub-40 ms time-to-first-token on TPUv5p hardware.
For developers, this means you can stream entire codebases, PDFs, or hours of video to a single endpoint and get accurate answers or generated code without extra retrieval logic. Built-in function calling lets the model browse the web, run code, and chain actions, making it easy to build agentic workflows—chatbots, data pipelines, or autonomous apps—via the Google AI Studio API.