OpenAI 發佈 GPT-6.1 Terra，主打日常電腦工作及較低成本

OpenAI on Monday released GPT-6.1 Terra, a faster and cheaper AI model built for the routine computer work that fills most office days. It replaces GPT-5.6 Terra and sits in the middle of OpenAI’s range, below Sol and Astra but above Luna.

The main change is what OpenAI calls computer use. Terra can look at a screen, understand what it shows, and then act on it by clicking, typing, and moving between apps, much as a person would.

In practice, that means a company could hand it a stream of incoming invoices. Terra would read each one and type the details into the accounting software. OpenAI’s other examples include filling in forms, drafting proposals, reading technical drawings and editing videos.

Terra is also better at understanding images, OpenAI said. In one of its tests, the model turned architectural drawings into a 3D model. In another, it watched an animated film shot by shot and wrote a time-coded log for a restoration team.

On that test, Terra scored 15 percentage points higher than the model it replaces, at about a third of the cost, according to OpenAI. The company said it also matched GPT-6 Astra, its most capable model, at a tenth of the price.

OpenAI also said Terra makes fewer factual mistakes. On a set of questions designed to trip models up, its error rate at the lowest setting fell from 53.6% to 16.7%.

For programmers, Terra solved about half the tasks in a test of coding at the command line, more than double its predecessor’s score. OpenAI still recommends Sol or Astra for the hardest coding work.

OpenAI compared Terra with Anthropic’s Claude Opus 5.5 too. It said Terra completed tasks at roughly a ninth of the cost, though these results come from OpenAI’s own testing and have not been independently checked.

The model is available from today to ChatGPT Plus, Pro, Enterprise and Edu users, and in Codex, OpenAI’s coding tool. Developers pay $1 per million input tokens and $5 per million output tokens, half the price of GPT-5.6 Terra.

On safety, OpenAI said Terra reaches the “Critical” level for cybersecurity skills under its Preparedness Framework, the system it uses to rate risky abilities. It carries the same protections as GPT-6 Astra.

The launch follows a setback last month, when OpenAI cancelled GPT-6.1 Astra after it failed internal safety tests. OpenAI said Terra is more likely than its predecessor to stay within the task it is given and to tell users about its limits.