Summary
OpenAI has introduced GPT-6.1 Terra, a model it says is designed for visual reasoning, computer use and coding at lower cost than its previous Terra model. It is available in ChatGPT Work and Codex for specified plans, and through the API.
OpenAI has introduced GPT-6.1 Terra, a model designed for visual reasoning and computer use as well as writing and coding. The company says it improves on GPT-5.6 Terra across several benchmarks while lowering standard API prices by 50% compared with GPT-5.6 Terra and GPT-6.1 Sol.
The announcement focuses on tasks in which an AI agent interprets visual information and then works with software—for example, reading an invoice and entering its details into accounting software. OpenAI reports gains in computer-use, visual-reasoning and coding evaluations, though the results described are the company’s benchmark findings.
Computer use, visual reasoning and speed
OpenAI tested computer-use performance on Agents’ Last Exam, a benchmark of professional tasks that includes turning architectural drawings into a 3D model and reviewing an animated film shot by shot to create a time-coded log. At medium reasoning effort, GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra at about one-third the cost, according to the company. OpenAI also says it performed on par with GPT-6 Astra at one-tenth the price per token.
At maximum reasoning effort, OpenAI reports that Terra was more than three times as token-efficient as Claude Opus 5.5. Combined with a price per token one-fourth as high, the company estimates roughly nine times lower cost per task. It also says Terra completed tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol in this evaluation.
On MMMU-Pro, a visual-reasoning benchmark, OpenAI says Terra outperformed GPT-5.6 Terra at every reasoning setting and nearly matched GPT-6 Astra and GPT-6.1 Sol. On OSWorld 2.1, which tests agents on multistep desktop-app tasks, Terra surpassed GPT-5.6 Terra’s best score at medium reasoning effort, at a lower cost per task.
Factuality and coding results
OpenAI reports lower error rates on its internal benchmark of prompts designed to induce 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, the company reports an 18-percentage-point decrease to 4.3%.
For software engineering, the model scored 50.3% on Terminal-Bench 4.0 at maximum reasoning effort. OpenAI compares that with the 21.5% published score for GPT-5.6 Terra. The company says GPT-6.1 Sol and GPT-6 Astra remain better choices for the most complex coding tasks.
On Terminal-Bench-Science 0.1, an evaluation of agents using code for scientific research tasks, OpenAI says Terra surpassed GPT-6.1 Sol at maximum reasoning effort and came within five points of Opus 5.5 at less than one-fifth of its estimated cost.
OpenAI says GPT-6.1 Terra meets the Critical threshold in cybersecurity under its Preparedness Framework and retains protections developed for GPT-6 Astra. The company also reports improvements in staying within task scope, disclosing limitations and resisting prompt-injection attempts.
Access and API pricing
GPT-6.1 Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Developers can access it through the API as gpt-6.1-terra.
Standard API prices are $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. The factuality figures come from OpenAI’s internal prompt benchmark, while the other performance results are tied to the named evaluations; they describe benchmark performance rather than a single general-purpose measure of model quality.