GPT-6.1 Terra Guide: Uses, Pricing and Sol Comparison

What is GPT-6.1 Terra useful for?

Visual tasks are one place to start. OpenAI lists matching product photos to catalog entries, interpreting technical drawings and understanding application screens as examples. When information is obscured or unclear, ask the model to identify uncertainty and check the result against the original image.

Terra also supports computer use. With a connected tool and the right permissions, it can interpret a screen and perform actions such as clicking or entering text. A workflow might read an invoice and enter its fields into an accounting application. Whether the full task succeeds depends on the tool, permissions, interface and required human approvals.

For writing and coding, Terra can draft, revise and perform bounded development tasks. OpenAI recommends GPT-6.1 Sol or GPT-6 Astra for the most complex coding tasks. These descriptions come from the launch announcement and model-selection guide.

Where can you use GPT-6.1 Terra?

As of the verification date, Terra is rolling out in ChatGPT Work and Codex for eligible Plus, Pro, Business, Enterprise and Edu users. Access depends on the rollout, plan, region and workspace settings. It is available in Work and Codex, rather than ordinary ChatGPT Chat. See the current release notes for details.

API developers can use the model ID gpt-6.1-terra. Its reasoning effort options are none, low, medium, high, xhigh and max; medium is the API default. Available controls can differ by product interface. See the model documentation.

GPT-6.1 Terra vs GPT-6.1 Sol: which should you choose?

Start with how readily you can check the task, then compare price.

Rates are standard API prices from the Terra and Sol model pages. They do not establish ChatGPT or Codex subscription usage deductions.

For a task with a fixed output format that is easy to check, Terra is a practical starting point. For changes spanning multiple modules, or where an error would require substantial manual repair, compare it with Sol before choosing a default. These are selection suggestions, not guarantees about every task.

How much does the Terra API cost?

Suppose you make 100 separate requests, each with 8,000 input tokens and 2,000 output tokens. The total is 800,000 input tokens and 200,000 output tokens. Assume standard pricing, no cache hits or cache writes, no image or tool charges, and each request below the long-context threshold:

• GPT-6.1 Terra: 800,000 input tokens cost $0.80, and 200,000 output tokens cost $1.00, for $1.80 total.
• GPT-6.1 Sol: the same tokens cost $1.60 for input and $2.00 for output, for $3.60 total.

This is a calculation under stated assumptions, not an actual bill. Retries, cached input, cache writes, images and tools can change the final amount. For Terra, the official pricing conditions say a request exceeding 272,000 input tokens uses twice the input and cached-input rates and 1.5 times the output rate for the full request. Fast mode has different rates.

For an application, measure the cost per accepted result as well as the price per token. A lower-rate model may need extra attempts or more manual corrections on a particular task.

What do the official benchmarks show?

OpenAI reports that GPT-6.1 Terra solved 50.3% of Terminal-Bench 4.0 tasks at maximum reasoning effort; the announcement lists GPT-5.6 Terra's published result as 21.5%. It also reports improvements on computer-use evaluations. These are OpenAI's reported results, not an independent test by this site.

Results depend on the task set, reasoning effort, tools and measurement conditions. A lower price per token in a benchmark does not establish a lower cost per completed task in every workflow.

How to try Terra on your own work

Choose a task where you can check the answer. For example, ask the model to extract dates, amounts and item names from an image of a table, and ask it to flag unreadable fields. Compare its output with the original.

For coding, give each model the same small issue with the same acceptance criteria. Record whether it understood the request, changed unrelated files, passed the existing tests and repaired any test failures. Use separate workspaces so the two runs do not interfere.

For computer-use tasks, review the intended actions. Keep a human approval step for sending messages, payments, deletion and permission changes. Use non-production data when practical, and check the final state. See OpenAI's computer-use guide.

Common questions

Is GPT-6.1 Terra free?

Availability in ChatGPT Work and Codex depends on the plan, rollout and workspace settings. API use is billed separately. Check your account and the current price page.

Can Terra accept audio or video directly?

The model page lists text and image input. Audio and video are not listed as native input modalities for this model. A video workflow may use extracted frames or other tools; visual understanding alone does not establish direct video-file support.

Can Terra use tools?

The model page lists tools including web search, file search, computer use and MCP. Availability depends on the integration. The documentation states that tool calling requires the Responses API.

Can Terra replace GPT-6.1 Sol for every task?

That depends on the task. Compare representative work using the same acceptance criteria and record quality, retries and review time. OpenAI continues to recommend Sol or Astra for the most complex coding tasks.

Choosing a starting point

GPT-6.1 Terra combines visual understanding, computer use and writing at a lower standard API rate than GPT-6.1 Sol. Try representative tasks, measure the complete workflow and compare the results with Sol before changing a production default. For the earlier model line-up, see the existing model-selection guide, noting its publication date.

Model and workflow guides