OpenAI on Monday, Oct. 26, released GPT-6.1 Terra, a model priced below its Sol and Astra models, which OpenAI says excels at the computer work that fills an office day: drafting proposals, reading technical drawings, filling out forms and editing videos. OpenAI calls it a significant step up from GPT-5.6 Terra, the version that came before it. It is also cheaper: OpenAI's standard prices for developers are half those of GPT-5.6 Terra and of its pricier GPT-6.1 Sol.
OpenAI says Terra is good at working a computer itself, the way AI agents do. AI agents are AI that can carry out tasks on their own, such as using apps and websites, instead of only answering questions. OpenAI says the lower price makes it practical to automate high-volume work, such as reading incoming invoices and typing their details into accounting software.
Terra is available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and in Codex, OpenAI's coding tool, and to developers through OpenAI's API, a connection that lets developers plug a service into their own apps. Developers pay by the token, the small chunks of text, often parts of words, that AI models read and write: at standard rates, $1 per million tokens sent to the model and $5 per million it writes back. OpenAI says the same ChatGPT plan allowance covers more tasks with Terra than with GPT-5.6 Terra, and far more than with its pricier Sol and Astra models. GitHub says Terra is also available in its Copilot coding assistant for paid plans, rolling out gradually.
On a test of professional computer tasks, OpenAI says Terra at its medium effort setting, one of several levels that set how hard the model works on a task, scores 15 percentage points higher than GPT-5.6 Terra at about a third of the cost. OpenAI says that on the same test it also performs on par with its top model, GPT-6 Astra, at one-tenth the price per token. OpenAI also says Terra finishes those tasks more than three times faster than Astra or Sol, and that at its highest effort setting a task costs about a ninth of what it costs with Anthropic's Claude Opus 5.5. On a test of fixing software problems by typing commands, such as repairing a webpage whose layout jumps as it loads, OpenAI says Terra at its highest effort setting solved 50.3% of tasks, up from 21.5% for GPT-5.6 Terra, though it says Sol and Astra remain the better choice for the hardest coding work.
OpenAI also says Terra makes fewer mistakes of fact. On its own set of requests designed to trip models up, the error rate at a low effort setting fell from 53.6% with GPT-5.6 Terra to 16.7%. In its safety report, OpenAI says it is treating Terra as reaching "Critical" capability in cybersecurity, a level it defines as a model that could find and exploit unknown security flaws in many well-protected critical systems without human help, and "High" in biology and chemistry, so Terra carries the same safeguards as GPT-6.1 Sol and GPT-6 Astra.
The test scores come from OpenAI's own runs. GitHub, which offers Terra in Copilot, says only that in its early testing the model solved noticeably more terminal and C# coding tasks than GPT-5.6 Terra. OpenAI's announcement does not say whether people on ChatGPT's free plan will get Terra.