OpenAI GPT-6.1 Terra and Google Gemini 4 Carbon Launch

OpenAI has released GPT-6.1 Terra, a model built to balance speed and cost on computer-based work. It is available through ChatGPT Work, Codex and the API, and supports drafting documents, interpreting images and taking actions in apps. API pricing is $1 per million input tokens and $5 per million output tokens.

Google answered with Gemini 4 Carbon, which targets long-running software development tasks and finding and fixing security flaws. The model scored 59 on Artificial Analysis Intelligence Index v5 and is opening first to selected cyber defense teams. No date for general access has been given. Starting API prices are $2 per million input tokens and $10 per million output tokens.

AI tools take on more everyday work

Gemini now lets users create, preview and edit documents, PDFs, presentations and spreadsheets directly in chat. Users can edit files themselves or select sections and ask Gemini to change them. PDF editing works on web and mobile, while other file types are limited to the web version. Google also opened a preview of its Connected Apps Platform, which lets third-party services plug into Gemini and Search; developers can test integrations in one place and submit them for review.

ChatGPT gained iCloud Mail and Calendar integrations for finding and managing email, drafting and sending messages, and editing events. They cover only iCloud services, not every account shown in Apple's Mail app. Apple said Siri AI will add Turkish, Traditional Chinese, Norwegian and Vietnamese by the end of the year. Meta launched Style Sketch, which builds a digital wardrobe from Instagram photos, and is available for now in Indonesia, Singapore and Thailand. Microsoft added more than 20 legal features to Copilot, including contract drafting and document review.

Other releases and science claims

Moonshot AI released Kimi K3.1 Pro and a faster Highspeed version, aimed at coding and generating code for Blender and CAD tools, and plans to publish the weights by November 4. Particle says its Particle-1, trained further from DeepSeek V4.1 Flash, uses about 45 percent fewer tokens at similar accuracy.

OpenAI says it obtained new results on three problems in mathematics, computer science and physics and is working with academics to review them. Anthropic says a research model produced work disproving the Exponential Time Hypothesis. Axiom said it formalized the full classification of finite simple groups in Lean, around 43 million lines of code produced in 24 days with human-guided AI agents.