OpenAI has introduced GPT-6.1 Terra, a new artificial intelligence model designed for computer use, visual reasoning, software development and professional tasks. The model joins the company’s GPT-6.1 family nearly a month after the announcement of GPT-6.1 Sol, offering improved performance over its predecessor while significantly reducing operating costs.
According to OpenAI, GPT-6.1 Terra can perform a wide range of computer-based activities, including completing browser forms, managing spreadsheets, interpreting technical drawings and editing videos. The company says the model can also complete computer tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol, making it a more affordable option for businesses and developers seeking to automate digital workflows.
The new model is also positioned as a cost-effective alternative to OpenAI’s more advanced systems. Although it doesn’t replace Astra for the most demanding workloads, its benchmark results suggest it can deliver competitive performance for many professional applications at substantially lower prices.
GPT-6.1 Terra delivers stronger results across professional tasks
OpenAI has reported notable improvements in GPT-6.1 Terra’s ability to handle complex computer-based work. On Agents’ Last Exam, a benchmark that evaluates how well AI agents complete professional tasks, the model scored 15 percentage points higher than GPT-5.6 Terra when operating at medium reasoning effort. OpenAI also reported that the cost per task was about one-third of its predecessor’s.
GPT-6.1 Terra: delivering fast and capable computer use at a lower price. GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it. https://t.co/8JTN1yonmV
— OpenAI (@OpenAI) October 26, 2026
The company said GPT-6.1 Terra performed on a par with GPT-6 Astra on the same benchmark while costing one-tenth as much per token. This result highlights the new model’s potential for organisations that need AI systems to complete multi-step tasks without incurring the higher costs associated with more powerful models.
Visual reasoning is another area where GPT-6.1 Terra has shown competitive results. On MMMU-Pro, a benchmark designed to assess an AI model’s ability to interpret and reason about visual information, Terra came close to the performance of both GPT-6 Astra and GPT-6.1 Sol.
These capabilities could make the model useful for tasks involving visual documents, technical diagrams and other information that requires more than straightforward text processing. Its ability to interact with computer interfaces also supports applications in which an AI agent must navigate software, interpret on-screen information, and carry out a series of actions to complete a task.
Software development performance has also improved significantly compared with GPT-5.6 Terra. At maximum reasoning effort, GPT-6.1 Terra scored 50.3 per cent on Terminal-Bench 4.0, up from 21.5 per cent for its predecessor. The benchmark measures an AI model’s ability to perform tasks in terminal-based development environments.
Despite the improvement, OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra for the most complex coding assignments. This suggests Terra is intended to balance performance, speed, and cost rather than serve as the strongest option for every development scenario.
New model significantly reduces API costs
GPT-6.1 Terra’s pricing is one of its main selling points. Through OpenAI’s application programming interface (API), the model costs US$1 per million input tokens and US$5 per million output tokens. Cached input tokens, which reduce costs when you reuse previously processed content, cost US$0.05 per million tokens.
By comparison, GPT-6 Astra costs US$10 per million input tokens and US$50 per million output tokens at standard API rates. This makes Terra approximately ten times cheaper for both input and output tokens, potentially offering considerable savings for developers running AI-powered applications at scale.
The lower prices could be particularly useful for companies that rely on AI agents to perform repetitive office tasks, process large amounts of information or interact with business software. For these applications, lowering the cost per request can make automation more practical, especially when workflows involve many separate actions.
However, token prices alone do not determine the total cost of completing a task. The number of model calls required, the length of the input and output, and the amount of reasoning needed can all affect overall spending. Performance differences between models may also influence how much additional processing is necessary to achieve a reliable result.
OpenAI is making GPT-6.1 Terra available to eligible paid users through ChatGPT Work and Codex, as well as via its API for developers. The rollout gives individual professionals and organisations more options to incorporate AI into their existing workflows.
ChatGPT Work provides access to AI assistance for professional activities, while Codex focuses on software development tasks. API access also lets developers integrate the model into their own products, services, and automated systems.
Terra targets businesses seeking affordable AI automation
GPT-6.1 Terra reflects a broader effort to make advanced AI capabilities more economical to deploy. Rather than requiring users to rely on the most expensive model for every activity, the new option offers a lower-cost alternative for workloads that do not necessarily demand the highest available reasoning performance.
For businesses, this distinction could help improve how AI resources are allocated. Terra could handle routine browser operations, spreadsheet updates, document interpretation, and other structured activities. At the same time, organisations can assign more demanding research, software engineering, and reasoning tasks to more capable models when needed.
Such an approach could help organisations balance operating costs with performance requirements. Developers building AI agents could also use Terra for high-volume operations, reserving more expensive models for situations that require deeper analysis or more sophisticated decision-making.
Nevertheless, OpenAI’s reported benchmark results do not establish that Terra will match Astra in every real-world situation. Benchmark scores measure performance under specific testing conditions, and real-world results may vary depending on task complexity, the software involved, and the required level of reliability.
The company’s recommendation to continue using GPT-6 Astra for its most demanding reasoning, coding, research, and professional workloads also distinguishes the models. Terra offers a more affordable route to computer-based automation, but Astra remains the preferred choice for tasks that require the highest level of capability.
With lower API prices, faster task completion, and improved coding results, GPT-6.1 Terra gives developers and businesses another option for deploying AI agents at scale. Its practical value will depend on whether the savings and performance improvements translate into reliable results across the applications that matter most to users.
