OpenAI Launches GPT-6.1 Terra for Faster, Lower-Cost Computer Use
OpenAI has introduced GPT-6.1 Terra, a new AI model designed to handle computer-based tasks with improved visual reasoning, more precise computer use, and lower operating costs. The model targets everyday professional workflows, including drafting proposals, interpreting technical drawings, completing forms, and editing videos.
Compared with GPT-5.6 Terra, the new model offers stronger performance across computer-use and coding benchmarks, alongside improvements in factual accuracy. OpenAI says it also delivers a more cost-efficient option for developers building AI agents that need to interact with desktop applications and complete tasks across multiple steps.
GPT-6.1 Terra Focuses on Computer Use
GPT-6.1 Terra combines visual understanding with the ability to take actions on a computer. Rather than simply interpreting screenshots or providing instructions, the model is designed to help carry out tasks that involve interacting with software, processing information, and completing work through a series of actions.
These capabilities could support workflows such as extracting information from invoices and entering it into accounting software, working with technical drawings, and handling repetitive office tasks. Its improved writing capabilities also make it suitable for communication-heavy work, including preparing proposals and other business documents.
OpenAI positions GPT-6.1 Terra as a balance between capability, speed, and cost. Its standard API pricing is 50% lower than that of both GPT-5.6 Terra and GPT-6.1 Sol, making it a potential option for businesses that need to automate large volumes of work without paying for a more expensive model on every task.
Computer-Use Benchmarks Show Performance Gains
OpenAI evaluated GPT-6.1 Terra on Agents’ Last Exam, a benchmark covering complex professional tasks. These include converting architectural drawings into 3D models and reviewing animated films shot by shot to produce time-coded logs for restoration teams.
At medium reasoning effort, GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra while costing approximately one-third as much to run. OpenAI also reports that the new model performs on par with GPT-6 Astra on this benchmark at just one-tenth of Astra’s price per token.
The comparison with Claude Opus 5.5 highlights the model’s potential efficiency at maximum reasoning effort. According to OpenAI, GPT-6.1 Terra is more than three times as token-efficient and costs one-quarter as much per token, resulting in an estimated ninefold reduction in the cost of completing a task. It also completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.
The model’s visual reasoning performance has also improved. On MMMU-Pro, a benchmark designed to evaluate visual understanding and reasoning, GPT-6.1 Terra outperforms GPT-5.6 Terra at every tested reasoning setting. OpenAI says it comes close to the scores of GPT-6 Astra and GPT-6.1 Sol while operating at a substantially lower cost.
Another evaluation, OSWorld 2.1, measures how well AI agents use desktop applications to complete multistep tasks. At medium reasoning effort, GPT-6.1 Terra exceeds the best score previously achieved by GPT-5.6 Terra, while requiring only a fraction of the cost per task.
Fewer Factual Errors and Better Coding Results
OpenAI has also focused on reducing hallucinations and factual mistakes. On its internal benchmark of prompts designed to trigger errors, GPT-6.1 Terra’s error rate fell from 53.6% with GPT-5.6 Terra to 16.7% at low reasoning effort, a reduction of nearly 37 percentage points.
At maximum reasoning effort, the error rate dropped by 18 percentage points to 4.3%, bringing the model close to GPT-6 Astra and GPT-6.1 Sol on this measure.
Software engineering is another area where GPT-6.1 Terra has made substantial progress. On Terminal-Bench 4.0, which tests coding agents on command-line problems such as fixing webpages with shifting layouts, the model solved 50.3% of tasks at maximum reasoning effort. That is more than double the published score of 21.5% for GPT-5.6 Terra.
However, OpenAI notes that GPT-6.1 Sol and GPT-6 Astra remain better choices for the most complex coding tasks.
GPT-6.1 Terra also performed strongly on Terminal-Bench-Science 0.1, which evaluates agents on scientific research tasks involving code. At maximum reasoning effort, it surpassed GPT-6.1 Sol at a lower cost and came within five percentage points of Claude Opus 5.5 while costing less than one-fifth as much by OpenAI’s estimate.
Safety and Availability
OpenAI says GPT-6.1 Terra retains the safeguards developed for GPT-6 Astra and meets the Critical threshold for cybersecurity under the company’s Preparedness Framework. Its alignment improvements are intended to help it stay within the scope of a user’s request, communicate limitations, and resist malicious prompt-injection instructions that attempt to redirect its actions.
The company says the model is available starting today to all Plus, Pro, Enterprise, and Edu users through ChatGPT Work and Codex. Developers can also access it through the OpenAI API using the model identifier gpt-6.1-terra.
GPT-6.1 Terra API Pricing
OpenAI’s standard API rates are:
- Input tokens: $1 per million tokens
- Cached input tokens: $0.05 per million tokens
- Output tokens: $5 per million tokens
With its combination of computer-use capabilities, improved factual accuracy, and lower pricing, GPT-6.1 Terra is positioned as a practical option for businesses and developers building AI agents for everyday work. More demanding coding tasks may still benefit from OpenAI’s higher-capability models, but Terra’s reported benchmark results suggest that many professional workflows could be automated at a significantly lower cost.
