OpenAI launches GPT-6.1 Terra for visual reasoning and computer tasks

The new model combines computer use with lower API pricing; OpenAI’s benchmark results remain vendor evaluations.

OpenAI released GPT-6.1 Terra on October 26, positioning the model for visual reasoning, computer interaction and everyday writing. Its launch announcement describes work such as interpreting technical drawings, entering invoice information and operating software through multistep tasks.

The company lists standard API prices of $1 per million input tokens, $0.05 for cached input and $5 for output. It says the model is available through the API and to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Those are published availability and pricing statements; individual account access still depends on the applicable product.

What the evaluation establishes

OpenAI reports improved computer-use and coding performance against its earlier Terra model. These are the vendor’s evaluations, rather than independent measurements of reliability in a particular organisation. A benchmark result cannot establish that an assistant will complete every invoice, drawing or editing workflow correctly.

More coding capability at a lower price

At maximum reasoning effort, OpenAI reports 50.3% on Terminal-Bench 4.0, compared with 21.5% for its published GPT-5.6 Terra result. The benchmark asks coding agents to carry out command-line tasks. Its launch also describes better performance on visual reasoning and desktop interaction tests.

This broadens the model’s proposed role beyond generating code snippets: reading an interface, deciding an action and checking the result all become part of the workflow. For developers comparing options, the important question is whether those improvements survive their own files and application states. The vendor figures make a case for evaluation; they do not replace a checked outcome.

Sources

- [Introducing GPT-6.1 Terra](https://openai.com/index/introducing-gpt-6-1-terra/) — OpenAI - [OpenAI API changelog — October 26](https://developers.openai.com/api/docs/changelog) — OpenAI