On October 26, 2026, OpenAI introduced GPT-6.1 Terra—a faster, more affordable model for computer use, image analysis, programming, and automation. Its standard API pricing is 50% lower than that of GPT-5.6 Terra and GPT-6.1 Sol. At the same time, the new model approaches the flagship GPT-6 Astra in some agent benchmarks. It is already available to developers through the API and is gradually rolling out to ChatGPT Work and Codex.
Key Takeaways
- GPT-6.1 Terra launched on October 26, 2026, with a focus on computer use, visual analysis, and practical AI agents.
- The API costs $1 per million input tokens and $5 per million output tokens; cached input costs $0.05.
- On Agents’ Last Exam, the new Terra outperforms GPT-5.6 Terra by 15 percentage points at medium reasoning effort.
- In Terminal-Bench 4.0, the model solved 50.3% of tasks, compared with 21.5% for GPT-5.6 Terra.
- The context window is 1.05 million tokens, with a maximum output of 128,000 tokens.
- Terra is available through the API, Work, and Codex, but not in standard ChatGPT chat.
GPT-6.1 Terra: What the New Model Can Do
OpenAI positions Terra as a cost-efficient model for tasks where AI needs not only to generate text but also to understand images, interact with interfaces, and carry out sequences of actions in software.
Examples include extracting data from invoices, entering information into accounting systems, analyzing technical drawings, filling out forms, working with product catalogs, and editing documents.
The model supports tool calls through the Responses API, structured outputs, web search, file handling, code execution, and computer use. It accepts text and images but does not support audio or video input as separate modalities.
How Much Does GPT-6.1 Terra Cost via the API?
OpenAI’s standard rates are:
| Token type | Price per 1 million |
|---|---|
| Input | $1.00 |
| Cached input | $0.05 |
| Output | $5.00 |
| 30-minute cache write | $1.25 |
For comparison, GPT-6.1 Sol and its predecessor GPT-5.6 Terra have standard input and output token rates that are twice as high. GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens—10 times more than Terra on these two measures.
For example, a task with 10,000 input tokens and 2,000 output tokens would cost approximately $0.02 on Terra, excluding caching, tools, and other surcharges. On Astra, the same token volumes would cost $0.20 at the base rates.
This does not automatically mean a tenfold saving on a completed task: different models may consume different numbers of tokens, tool calls, and retries.
Separate terms apply to long requests. If the input exceeds 272,000 tokens, higher rates apply to the entire request. Batch and Flex modes, by contrast, offer a 50% discount on standard pricing.
How Does GPT-6.1 Terra Compare with Flagship Models?
OpenAI has published results from several benchmarks showing substantial improvement over GPT-5.6 Terra. However, these are primarily measurements conducted by the company itself, not independent verification of every claimed advantage.
Computer Use
On Agents’ Last Exam at medium reasoning effort, GPT-6.1 Terra outperformed the previous Terra by 15 percentage points at roughly one-third the execution cost.
According to OpenAI, the new model reaches GPT-6 Astra’s level on this benchmark at approximately one-tenth the token price. The company also reports that Terra completes the tested tasks more than three times faster than Astra and GPT-6.1 Sol.
These results relate to a specific set of agent tasks. They do not establish equal model quality in advanced programming, scientific research, or extended reasoning.
Programming and Factual Accuracy
In Terminal-Bench 4.0 at maximum reasoning effort, GPT-6.1 Terra solved 50.3% of tasks. The published result for GPT-5.6 Terra was 21.5%.
OpenAI also reports a lower rate of factual errors on its own set of difficult queries. At low reasoning effort, the rate fell from 53.6% to 16.7%, while at maximum effort it dropped to 4.3%.
These are internal test results, so they should not be interpreted as a universal error rate for the model in real-world use.
For the most demanding programming tasks, OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra. UAGEEK previously took an in-depth look at GPT-6 Astra’s capabilities, pricing, and agent benchmarks.
Where Is GPT-6.1 Terra Available?
Developers can use the model through the OpenAI API under the identifier gpt-6.1-terra. It supports the Responses API and Chat Completions API, although tool calls require the Responses API.
Across OpenAI products, the model is gradually rolling out in ChatGPT Work and Codex for eligible plans and workspaces. The company’s release notes mention Plus, Pro, Business, Enterprise, and Edu, but actual access depends on the plan, region, and administrator settings.
Important clarification: GPT-6.1 Terra is not available as a model in ChatGPT’s regular Chat mode. Therefore, saying that it is ‘available in ChatGPT for all users’ would be inaccurate.
What Terra Changes for Business and Automation
Terra’s most interesting use case is automating large volumes of repetitive tasks for which a flagship model would be too expensive.
For marketing teams, this could include drafting reports, checking spreadsheet structures, transferring data between systems, classifying advertising materials, and working with analytics platform interfaces.
However, the ability to control a computer does not guarantee reliable execution of every workflow. Actions involving advertising budgets, payments, campaign publishing, and changes to access permissions require separate authorizations, logging, and human approval.
For businesses, it makes more sense to compare models by the cost of a successfully completed task rather than token rates alone. Execution time, error counts, retries, and the cost of reviewing results must all be considered.
Why a Cheaper Terra Is Not Necessarily a Weaker Model
GPT-6.1 Terra occupies a more economical position in OpenAI’s lineup, but it would be inaccurate to describe it simply as a stripped-down version of the flagship. The model is optimized for a particular balance of speed, cost, and quality, and in some agent benchmarks it delivers results close to those of far more expensive systems.
The release shows that competition between AI models increasingly depends not on maximum raw capability, but on how much it costs to automate real work. In this segment, Terra may prove more attractive than pricier models.
Sources for Verification
Sources for Verification
OpenAI — Introducing GPT-6.1 Terra. Official announcement dated October 26: the model’s intended use, comparison with GPT-5.6 Terra, agent benchmarks, pricing, and availability.
Official GPT-6.1 Terra announcement
OpenAI API — GPT-6.1 Terra Model. Technical specifications, context window, supported tools, reasoning modes, and detailed pricing.
GPT-6.1 Terra model documentation
OpenAI — Release Notes. Confirmation of the October 26 launch, gradual rollout in Work and Codex, and Terra’s absence from regular Chat.
OpenAI release notes
OpenAI API — Changelog. Confirmation of the gpt-6.1-terra model release on October 26 and API availability.
OpenAI API changelog
OpenAI Deployment Safety. System documentation on risk assessments and safety mitigations for GPT-6.1 Terra.
OpenAI model safety documentation
