GPT-6.1 Terra Released: OpenAI Leads with Computer Use, Commenters Ask About GPT-7 First

OpenAI has released GPT-6.1 Terra.

The official keywords are clear: lower price, faster speed, and the ability to directly operate a computer. Sam Altman says it is good at many things, but what OpenAI is especially excited about is the new stage of computer use.

According to the official description, Terra can understand what is on the screen, execute actions across applications, write naturally, is suitable for a full workday, and is efficient enough to take on more tasks. For ordinary users, computer use can be understood as the model looking at the screen, clicking buttons, filling out forms, and moving back and forth between software.

In official benchmarks, Terra significantly outperforms GPT-5.6 Terra on Agents’ Last Exam. This test measures an agent's ability to use software at work. OpenAI also claims it matches GPT-6 Astra, costs only one-tenth as much per token, and completes tasks more than three times faster on average than Astra and GPT-6.1 Sol.

Availability has also been announced. Terra is rolling out to Plus, Pro, Enterprise, and Edu users, covering ChatGPT Work and Codex. Developers can use gpt-6.1-terra via the OpenAI API. The official announcement is here, and the launch tweet is here.

The comments are more informative than the launch copy. Under Sam's tweet, someone directly asks about GPT-7. Someone says they used it for an hour. Someone complains that one of Terra's headline features is using a computer, and the bar for an "era" is really low. Another writes that nobody cares about faster and cheaper; we need a better model. Another reply is even more direct: we need a better model, not a worse one.

There are also practical-minded people. A developer suggests that ChatGPT Desktop allow presetting the computer use model. For example, let Astra handle planning and judgment, and let Terra click buttons and go through workflows. This division of labor is closer to a real product than one model doing everything.

One commenter says the computer use benchmark he really cares about is canceling a gym membership and not being transferred to the retention department. This example is very concrete. Operations in demo videos are usually smooth, but the trouble with everyday software lies in logins, pop-ups, CAPTCHAs, canceling subscriptions, customer service scripts, and cross-app state. Cheaper and faster models can let agents run a few more steps, but in the end users remember only one thing: whether it got the job done.

Someone posted a moon landing video comparing Opus and Terra side by side.

Terra's positioning is clear: drive down the cost and speed of agents so that computer use can be rolled out more easily. The split in the comments is also clear: some users want the upper limit of capability, while others want a price that can get work done. Both needs exist at the same time. Terra's test is not in the launch video, but in those workflows that nobody wants to record in a demo.