Next-Gen AI Models and Persistent Agents Arrive from Major Providers

OpenAI announced GPT-6.1 Sol, which offers near-Astra intelligence at two dollars per million input tokens and ten dollars per million output tokens. The company also introduced Dots, an always-on AI agent powered by GPT-6 Astra that runs background tasks 24/7 on its own cloud computer. A new five hundred dollar monthly Pro subscription provides access to an Ultrafast mode that generates up to three hundred tokens per second in Codex.

Anthropic released Claude Sonnet 5.5, delivering over thirty percent faster generation speeds and reduced token costs. The model is priced at two dollars per million input tokens and ten dollars per million output tokens, with cache reads costing twenty cents and cache writes costing two dollars and fifty cents. It scored 70.6 percent on the Terminal-Bench 4.0 agentic coding benchmark, outperforming Claude Opus 5.5.

Google announced Gemini 4 Argon, a new flagship model optimized for complex workflows, software engineering, and cybersecurity. The model features an expanded output limit of one million tokens and is priced at two dollars per million input tokens and ten dollars per million output tokens. It achieved a 77.9 percent score on the DeepSWE v1.1 software engineering benchmark and a 68 percent score on CWE-bench v1 for cybersecurity.

TypeSafe AI launched Jev, a specialized non-writing decision model that outputs probabilities and scores instead of generating text. Jev is designed for categorical and probabilistic decisions, running up to two hundred times faster and four hundred times cheaper than standard large language models. It costs approximately four cents per million input tokens on OpenRouter and answers choice, score, and binary questions.

Manus 2.0 launched with the Cascade engine, which runs over twenty-eight percent faster and uses over twenty-three percent fewer tokens than its predecessor. The update introduces persistent Cloud Computers, which are always-on virtual machines that allow agents to run automations even when the user's computer is closed. It also features Cue, a standalone application for managing personal AI agents that receive their own email addresses, phone numbers, and cloud environments.

Microsoft upgraded Microsoft 365 Copilot, structuring it into three core sections: Home, Code, and Autopilot. The Code section allows users to build software and interactive dashboards from plain English descriptions. Autopilot acts as an autonomous background agent that takes on goals, tracks progress, and executes tasks across Microsoft 365 applications without active user supervision.