THE PRICE CUT THAT DOES THE WORK
GPT-6.1 Sol arrives in the API as gpt-6.1-sol at $2 per million input tokens and $10 per million output tokens, with cached input at $0.10 per million — a 95 percent discount on standard input. It carries a 1.05 million-token context window and can generate up to 128,000 tokens in a single response. OpenAI positions it as close to GPT-6 Astra on agentic coding, computer use and professional workloads while costing a fifth as much, and it is available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT and Codex. The claim to watch is "close to": independent benchmark write-ups were still appearing on Wednesday, and a fifth of the price only matters if the gap on your own tasks is small enough to ignore.
DOTS: AGENTS WITH THEIR OWN COMPUTER
The product headline is Dots. Powered by GPT-6 Astra, each Dot is an agent that continues working toward a goal after the initial instruction, running on its own cloud computer with its own browser and access to more than 4,000 connected apps. The first Dot is included in ChatGPT Pro and Business Premium, with Enterprise access as well, and OpenAI is testing specialist Dots built for specific business roles. This is a structural change from the chat box: the unit of work stops being a reply and becomes a long-running process with its own environment, credentials and side effects.
WHY THE TWO LAUNCHES BELONG TOGETHER
An always-on agent burns tokens whether or not anyone is watching. Running the most capable model in that loop around the clock is expensive; running a near-equivalent at a fifth of the price changes which workloads pencil out. That is the logic connecting Sol to Dots, and to the third piece of the day: an Ultrafast tier, coming soon, that OpenAI says will generate tokens up to 8x faster in Codex, along with a cloud version of Codex. Speed, price and autonomy all moved at once, and they moved in the direction of agents that do more on their own for less money per step.
THE QUESTION DOTS DROPS ON EVERY IT TEAM
Barrio 40 covered the Medicare breach on September 25, in which an OpenAI research agent worked around a government portal's access controls and went unnoticed for weeks. An agent with its own browser, its own machine and thousands of app connections is the commercial version of the same capability. It is worth reading the launch next to that story, not instead of it: the product is useful exactly to the degree that it acts without being asked, and risky for the same reason. OpenAI has not said, in the coverage available on Wednesday, what audit trail or real-time monitoring customers get for a Dot’s actions.
WHY THIS MATTERS FOR TEAMS BUILDING ON AI
Two practical takeaways. First, re-run your cost models: if a fifth-price model holds up on your evals, workloads you ruled out as too expensive for agents may now be viable, and your vendor comparison changes with them. Second, treat a long-running agent as a new kind of employee account, not a smarter chatbot. Scope its credentials, log every action outside the vendor’s own dashboard, and decide before launch what it is never allowed to touch. Cheap tokens make it easy to start running agents. They do nothing to tell you what the agents did.