Gemini Carbon API out today (aistudio.google.com)

Google has launched the Gemini 4 Carbon API, heralded as the company's most powerful reasoning model, designed to support complex workflows in software engineering, scientific reasoning, and cybersecurity. Unlike its predecessor, Gemini 4 Argon, Carbon prioritizes deeper reasoning by utilizing more "thinking tokens," enhancing its ability to tackle intricate problems and execute rigorous build-and-verify loops. This model offers a unified context window of 1 million tokens and is capable of producing longer responses, making it particularly useful for demanding tasks involving multimodal knowledge and detailed data analysis. The significance of Gemini 4 Carbon lies in its advanced capabilities, such as improved precision on demanding knowledge work and a boost in performance for scientific proofs and vulnerability discovery. It caters to users needing higher reasoning depth, ensuring greater accuracy by grounding sources effectively in dense datasets. The model's tiered "thinking level" feature allows for adjustable reasoning efforts, making it flexible for varied use cases while keeping costs manageable—introductory pricing is set at $2/1M input tokens and $10/1M output tokens until November 2026. Overall, the introduction of this model represents a significant leap forward in the AI/ML landscape, enhancing both the development and debugging capacities for sophisticated AI-driven applications.