Gemini 4 Carbon announced, and Gemini now keeps requests

Google announced Gemini 4 Carbon on 28 October, its strongest model, built on Gemini 4 Argon, but only a subset of cyber defenders get it at first, then more trusted partners through its Fairwind Program. Developers get it later, on no date yet, at Argon's launch price of $2 in and $10 out per million tokens, then $4 and $20. Google reports 0.0% of attacks getting through on Gray Swan's prompt injection benchmark, against 0.7% for Argon. It is not in the Gemini API yet, so keep building on Argon; Gemini 4 Carbon covers what Google has and has not said.

Google released the Gemini API CLI in beta on 27 October: gemini-api, a single command that an agent with shell access can run to generate text, images, speech and video, analyse files, or start a managed agent such as Deep Research in the background, with no SDK code in the project. It reads GEMINI_API_KEY. Google says breaking changes can come between minor versions, so pin a release in scripts and CI.

Since 26 October, a new paid-tier project's Generate Content API calls are stored by default (store=true), as Interactions API calls already were. Stored logs expire after 55 days unless you set 7, 14 or 28, and logs you save to a dataset do not expire. Projects created earlier keep their setting. If you send personal or customer data, set store=false on each request or turn logging off in the Logs and Datasets settings of each new project.

OpenAI released gpt-6.1-terra on 26 October, a model for visual reasoning, precise computer use and natural writing, at $1 in and $5 out per million tokens: half of GPT-6.1 Sol and 10 times GPT-6 Luna. It takes text and images, has a 1,050,000-token window and writes up to 128,000 tokens, with reasoning effort from none to max. It is also in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. If Sol is more than a job needs and Luna is not enough, test Terra in between; GPT-6 Astra vs Sol vs Luna shows where it fits.