OpenAI released GPT-6.1 Terra on October 26, 2026, adding a lower-priced option for workers and businesses using AI to handle documents and applications.
OpenAI positions GPT-6.1 Terra for writing, images and application screens. Its suggested uses include preparing proposals, interpreting charts and filling forms from source documents. These are the company’s recommendations, not evidence that every task will be cheaper or more accurate.
Who can use Terra
The product changelog lists Terra in ChatGPT Work and Codex on Plus, Pro, Business, Enterprise and Edu plans—not ordinary Chat. On desktop and web, users can select it through Advanced; Work also supports it on mobile. Enterprise access depends on workspace and administrator controls. UAE inference residency—keeping model execution in the UAE—is not supported.
For subscribers, Work and Codex share usage allowances. API prices apply separately to software accessing the model through OpenAI’s metered service. They do not tell subscribers how many tasks their plan includes: context, tools and task complexity affect usage.
What the lower price buys
Tokens are the units used to count material a model processes and generates. For Standard API requests with no more than 272,000 input tokens, Terra costs US$1 per million uncached input tokens and $5 per million output tokens. Sol’s equivalent rates are $2 and $10. Cached input has a separate, lower rate.
An illustrative calculation: assume 1,000 requests, each using exactly 10,000 uncached input tokens and 2,000 total billable output tokens. At those Standard rates, Terra’s model charges would be $20—$10 for input plus $10 for output—against $40 for Sol. That assumes identical token volumes and equally usable results, neither of which is established by the price list. It excludes tools, retries, infrastructure, taxes and human review.
Large document bundles can cost more. Above 272,000 input tokens, Terra doubles input and cache rates and charges 1.5 times the output rate for the entire request, not just the excess. Fast mode also costs twice Standard. Compare the same processing mode and request size before budgeting.
Where to start
For a first comparison, choose a familiar proposal, chart or form-filling task and define what a correct result must contain. OpenAI’s model guide still recommends Luna for focused, repeatable jobs and Astra for the most demanding work. It also advises checking names, numbers and visual details against the original material.
For tasks that act inside applications, OpenAI’s computer-use guidance calls for limited access, user control over consequential actions and verification of the actual outcome. A lower token price does not remove those requirements.
The purchasing question is the cost of work you can accept, including corrections—not merely the cost of a response. For a separate example of that distinction, see RohitAI’s analysis of Asana’s browser-agent cost study; that study did not evaluate Terra.