OpenAI launched GPT-6.1 Terra on Monday, a faster and cheaper model for routine computer work. Openai 6.1 Terra can look at a screen, understand what it shows, and act on it by clicking, typing, and moving between apps.
For developers, the model costs $1 per million input tokens and $5 per million output tokens. That is half the price of GPT-5.6 Terra, while OpenAI also says Terra can match GPT-6 Astra at a tenth of the price.
GPT-6.1 Terra in ChatGPT
The model is available to ChatGPT Plus, Pro, Enterprise and Edu users, and in Codex. OpenAI is aiming it at invoice entry, form filling, proposal drafting, technical drawings, and video editing, which means the model is being pushed into repetitive office tasks rather than only chat-style use.
OpenAI also says Terra can read incoming invoices and type the details into accounting software. That narrows the gap between a model that answers prompts and one that can handle a basic workflow from screen to software.
OpenAI pricing and comparisons
OpenAI says Terra replaces GPT-5.6 Terra and sits below Sol and Astra but above Luna in its range. The company says Terra turned architectural drawings into a 3D model in one test, and watched an animated film shot by shot to write a time-coded log for a restoration team in another.
On that film task, Terra scored 15 percentage points higher than the model it replaces, at about a third of the cost. For coding at the command line, OpenAI says Terra solved about half the tasks and posted a score more than double its predecessor’s.
OpenAI safety and limits
The company says Terra makes fewer factual mistakes, with its error rate at the lowest setting falling from 53.6% to 16.7% on questions built to trip models up. OpenAI also says Terra is more likely than its predecessor to stay within the task it is given and to tell users about its limits.
That sits beside a sharper concern: OpenAI says Terra reaches the Critical level for cybersecurity skills under its Preparedness Framework, while still carrying the same protections as GPT-6 Astra. The open question is how those protections work in practice for users who will now be able to point the model at screens, files, and live apps.
