Early GPT-6.1 Terra tests point to Astra-class computer use at a tenth of the price

Early hands-on reports of GPT-6.1 Terra are circulating, and the two themes that dominate them are price and computer operation, according to QbitAI.

By the report's account, users are getting Astra-class computer-operation capability at roughly one-tenth of the price. That capability means the model acts inside a graphical interface, clicking, typing and navigating, rather than only returning text.

Computer use is treated as a threshold because it turns a model from an adviser into an executor. The model has to read the screen, plan multi-step actions and recover from errors, and a failure at any step stops the task.

Price is the other half of the story. A tenth of the cost per task changes the economics of automation that depends on many repeated calls — form filling, bulk data entry, chaining workflows across applications.

The same day, OpenAI also published three new AI science findings, the report notes. A product release and research results landing in the same window suggest the company is pushing availability and capability-boundary narratives at once.

The caveat is that first-wave user tests are early samples. Test environments, task types and success-rate accounting are not standardised, so Astra-class reads more like a description of a capability band than a reproducible benchmark score.

What developers actually need to verify is stability over long task chains: whether a run is still controllable after twenty steps, how the model handles pop-ups, CAPTCHAs and permission prompts, and whether failures leave an auditable log.

Next to watch is whether official pricing and rate limits match the testers' numbers, and where computer operation shows up in real enterprise workflows. If the cost curve has moved as reported, the business model for agent products moves with it.