VarwoOpenAI GPT-6.1 Terra: Access, API Pricing and When to Try It
GPT-6.1 Terra is positioned for visual reasoning, computer use and clear writing. Choose the appropriate Work, Codex or API route, then evaluate one bounded task before adopting it for a larger workflow.
Launch announcement · Official model specifications · API pricing · Work and Codex availability
Official OpenAI documentation checked October 27, 2026. This guide explains access and billing using the linked provider sources. We did not run the model, call a paid API or benchmark computer use.
What is GPT-6.1 Terra?
The official model page identifies gpt-6.1-terra as a model for visual reasoning, computer use and writing. It lists a 1,050,000-token context window, up to 128,000 output tokens, text and image input, and text output. Native audio and video modalities are not listed as supported.
Those specifications describe capacity and interfaces. They do not establish that a model will correctly interpret every chart, click the right control or finish a particular document. Evaluate the final work against your sources rather than relying on the size of the context window.
Choose the correct access route
The official Work announcement lists Plus, Pro, Business, Enterprise and Edu access through Work and Codex. It describes selecting Terra in Advanced, with workspace controls affecting enterprise availability. It explicitly distinguishes these surfaces from Chat, where the model is not available, and excludes UAE inference residency.
For an application integration, use the API model ID and verify access in the intended project. For tool calling, the model documentation requires the Responses API. It lists reasoning.effort values none, low, medium, high, xhigh and max, with medium as the default. We did not verify access for any particular account.
API prices and the long-context boundary
The official pricing table gives these Standard, short-context USD rates per one million tokens:
| Usage category | Listed rate |
|---|---|
| Input | $1.00 |
| Output | $5.00 |
| Cached input | $0.05 |
| Cache write, 30-minute retention | $1.25 |
| Cache write, 12-hour retention | $2.50 |
The model page's pricing notes say that prompts with more than 272,000 input tokens apply 2× input and cache rates and 1.5× output rates to the full request. Do not apply the higher rate only to the portion above that boundary. Different service modes, regional processing and tool usage can change the price; check the billing route actually selected.
Two illustrative budgets
These are arithmetic examples, not measured bills. Both assume Standard mode, no caching, no tools and no other charges.
| Reported usage | Input calculation | Output calculation | Illustrative total |
|---|---|---|---|
| 10,000 input + 2,000 output tokens | 10,000 ÷ 1,000,000 × $1 = $0.010 | 2,000 ÷ 1,000,000 × $5 = $0.010 | $0.020 |
| 300,000 input + 2,000 output tokens | 300,000 ÷ 1,000,000 × $2 = $0.600 | 2,000 ÷ 1,000,000 × $7.50 = $0.015 | $0.615 |
The second example crosses the long-context threshold. Use the provider's complete reported usage; the visible answer length alone is not a billing record. These API figures also do not quote the cost or usage allowance of a Work or Codex subscription.
Evaluate one useful task
The following is an original, untested evaluation plan, not a report of Terra's performance.
- Fix the intended result. For a document, choose a short source and define the required points. For an image, identify the labels or relationships that must be read correctly.
- Specify evidence. Ask for references to the supplied sections or visible image regions. Include a question whose answer is deliberately absent, and check whether the model acknowledges the gap.
- Bound any action. For a computer-use task, define the allowed destination and the observable stopping point. Review a consequential action before carrying it out.
- Inspect the result. Check names, amounts, citations and formatting against the original material. For an app task, inspect the resulting state rather than accepting a narrative that the task succeeded.
- Record settings and usage. Keep the model ID, effort setting, task size and observed failures. Compare another setting only when a specific problem warrants it.
That process gives you a small acceptance sample. It does not establish broad accuracy, production reliability or a universal preference over another model.
Sources and related guides
- OpenAI: GPT-6.1 Terra launch announcement
- OpenAI: GPT-6.1 Terra model documentation
- OpenAI: API pricing
- ChatGPT Learn: Terra in Work and Codex
- ChatGPT tool overview
- Kimi K3.1 Pro API cost guide
Prices and availability are limited to the official pages reviewed on October 27, 2026. Check the current provider pages before committing a budget.