Summary
OpenAI has released GPT-6.1 Terra for ChatGPT Work, Codex and its API. The company reports gains in computer-use, coding and factuality benchmarks, alongside lower standard API prices than for GPT-5.6 Terra and GPT-6.1 Sol.
OpenAI has introduced GPT-6.1 Terra, a model designed to combine visual reasoning with computer use and text generation. It is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and to developers through the API as gpt-6.1-terra.
The company positions Terra as a faster, lower-cost option for work involving desktop applications, such as interpreting technical drawings, filling forms and editing videos. OpenAI says its standard API prices are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol. The listed rates are $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens.
Visual reasoning and computer use
Computer-use models are evaluated on tasks that require both interpreting what is on a screen and taking steps in software. OpenAI reports that, at medium reasoning effort on Agents’ Last Exam—a benchmark of professional tasks—GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra at about one-third the cost per task. The company says Terra performed on par with GPT-6 Astra on this benchmark at one-tenth its price per token.
On MMMU-Pro, a visual-reasoning benchmark, OpenAI reports that Terra outperformed GPT-5.6 Terra at every tested reasoning setting and came close to scores for GPT-6 Astra and GPT-6.1 Sol. On OSWorld 2.1, which tests multistep tasks using desktop applications, Terra at medium reasoning effort exceeded GPT-5.6 Terra’s best score at a fraction of the cost per task.
These results matter for developers considering AI agents for repetitive, screen-based work: the relevant trade-off is not only whether a model can complete a task, but also how quickly and economically it can do so. OpenAI says Terra completed tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol in its reported comparisons.
Reported changes in factuality and coding
OpenAI also reports lower error rates on an internal benchmark made up of prompts designed to elicit errors. At low reasoning effort, the rate fell from 53.6% for GPT-5.6 Terra to 16.7% for GPT-6.1 Terra. At maximum effort, it fell by 18 percentage points to 4.3%. These figures describe performance on that specific internal test.
For software engineering, GPT-6.1 Terra solved 50.3% of tasks in Terminal-Bench 4.0 at maximum reasoning effort, compared with GPT-5.6 Terra’s published score of 21.5%. OpenAI says GPT-6.1 Sol and GPT-6 Astra remain better suited to its most complex coding tasks. In the research-focused Terminal-Bench-Science 0.1, the company reports that Terra surpassed GPT-6.1 Sol at maximum reasoning effort and came within five points of Claude Opus 5.5 at less than one-fifth the estimated cost.
OpenAI says the model retains the protections developed for GPT-6 Astra and meets the Critical threshold in cybersecurity under its Preparedness Framework. The benchmark results are company-reported; scores on defined tests indicate performance in those test settings, rather than guaranteeing the same results across every real-world workflow.