OpenAI Releases GPT-6.1 Terra: Astra-Level Computer Use at 1/10 the Price

OpenAI has released GPT-6.1 Terra, a model meant to make computer control cheap and fast. The company claims Terra matches Astra in its ability to work with interfaces, while costing roughly ten times less. The release came on October 27, the day Anthropic filed for an IPO. The timing is telling: as one of its main competitors heads for the public market, the other is pushing on price.

A bet on cheap computer use

OpenAI describes Terra as a model that delivers "fast, strong computer use for less money." It combines three things: detailed visual reasoning, precise actions in the interface, and clearer, more natural writing.

The first two are essential for any agent that clicks, fills in forms, and navigates websites on its own. The model has to work out what is on the screen: a button, an input field, a pop-up, or a banner. Then it has to land on the right spot and stay on track through a long chain of steps. An error at step five usually ruins the entire workflow, so accuracy matters more here than polished answers in a chat.

The third point is practical too. An agent does not just press buttons. It also writes emails, reports, and comments. If the text sounds unnatural, a person has to rewrite it, and the time savings disappear.

Why price matters most

The headline figure of the release is 1/10 of Astra's cost. Agentic tasks are expensive by nature. A single operation like "find a supplier, check prices, fill in a request" takes dozens of model calls, and at every step the model parses a screenshot again. At high prices, such scenarios remain demos. With a tenfold price cut, they become routine automation.

If independent tests confirm the claimed parity with Astra, it will change the math for companies building agents on top of an API:

• routine office workflows, interface testing, and work with legacy software that has no API get a lower barrier to entry;
• top-tier expensive models will have to be justified only by the tasks Terra cannot handle;
• developers of agent platforms can more easily build setups in which a cheap model does most of the work and an expensive one steps in for hard cases.

What we know about the release

• Model: GPT-6.1 Terra, developed by OpenAI.
• Positioning: fast, strong computer use at a lower price.
• Key capabilities: detailed visual reasoning, precise computer control, clearer and more natural writing.
• Comparison benchmark: Astra — Terra is described as comparable in computer use capabilities.
• Price: roughly 1/10 of Astra's cost.
• Public reaction: Sam Altman responded to the release with clear enthusiasm.
• Context of the day: Anthropic filed for an IPO.

Background: a release against the backdrop of

Anthropic's IPO

For Anthropic, going public means a new level of disclosure: revenue, margins, compute costs, and dependence on a handful of large customers will all become subjects of investor analysis. In that market, price pressure from OpenAI becomes part of the investment story itself. If comparable quality on agentic tasks can be bought for ten times less, it affects margin expectations for every provider of frontier models.

OpenAI, meanwhile, is showing that it competes not only on model size but also on the economics of use. For businesses, the question is now a practical one: how much does one successfully completed workflow on a real computer cost?

What to check before deployment

Teams planning to use Terra should run their own comparison on their own tasks. Take a set of 50–100 real scenarios: CRM requests, exports from internal systems, form filling. Run them on Terra and on your current model, then compare the share of successfully completed tasks, execution time, and total cost.

Separately, test behavior on non-standard screens: captchas, layout changes, unexpected dialogs. That is where agentic models break most often. If Terra confirms the claimed level, it makes sense to put it on high-volume routine operations and keep the expensive models for rare, complex cases.