How Much To Run AI
Hosted API

GPT-6.1 Terra: Computer Use, API Setup and Task Cost

GPT-6.1 Terra is OpenAI’s model for visual reasoning, computer use and everyday writing. It reads text and images and produces text or tool calls. Use it in ChatGPT Work or Codex, or connect the API to an application and an execution environment. Standard API rates start at $1 per million input tokens and $5 per million output tokens; a completed task can include many model turns, screenshots and retries.

Content reviewed:

GPT-6.1 Terra computer-use loop: observe a screenshot, execute model-selected actions, and verify the result; complete task cost includes every turn, tools and the execution environment.
Computer-use workflow and cost boundaries. This is an explanatory diagram, not a recorded model run.

Use this model

API docs and console

Use cases

  • Fill a form or navigate a desktop application with a defined completion check.
  • Read a screenshot or technical drawing and produce a written answer.
  • Compare the cost of Terra and Sol on the same accepted task.

Where can I use GPT-6.1 Terra?

OpenAI announced Terra on October 26, 2026, for ChatGPT Work and Codex on Plus, Pro, Enterprise and Edu plans, and for developers through the API. A signed-in product uses its plan allowance; API requests use separate token billing.

For an integration, select gpt-6.1-terra. The Responses API supports tools; Chat Completions supports requests without tool calling. Start at the default medium reasoning effort. The supported levels are none, low, medium, high, xhigh and max.

Connect computer use to a real application

The model needs a tool that can observe and operate the target environment. OpenAI documents a structured computer tool and code-driven desktop integrations. Your application executes the returned actions and sends the resulting screen back; declaring a tool alone does not complete the task.

Start in a disposable desktop or test account. Give the agent a specific goal and a way to check it, such as saving a form and checking the saved values. Read external page content as task data, and keep approval for purchases, messages and destructive actions in your application.

For a first API check, send a small text request before adding the desktop driver. Then follow the computer-use guide to implement the action-and-observation loop. Keep the API credential on the server.

Minimal Responses API request; this does not operate a desktop
from openai import OpenAI

client = OpenAI()
response = client.responses.create(
    model="gpt-6.1-terra",
    reasoning={"effort": "medium"},
    input="Write a checklist for verifying the saved values of a web form."
)
print(response.output_text)

GPT-6.1 Terra pricing: budget the full task

At Standard rates, 50,000 fresh input tokens and 5,000 output tokens cost $0.075: $0.05 input plus $0.025 output. This assumed request excludes cache writes, tool charges and the execution environment. A ten-turn workflow needs the usage from all ten turns, including repeated screenshots and tool results.

Cache reads cost $0.05 per million tokens. Explicit cache writes cost $1.25 for 30-minute retention or $2.50 for 12-hour retention. Above 272,000 input tokens, the full request uses twice the input and cache rates and 1.5 times the output rate. Thus 300,000 fresh input tokens plus 10,000 output tokens cost $0.675.

Fast mode doubles Standard rates. Batch and Flex are 50% lower, while regional processing adds 10% where available. Choose a mode that fits the task’s latency and tool requirements, then reconcile the provider’s usage with the bill.

Standard itemUSD per 1M tokens
Fresh input$1.00
Cached input reads$0.05
30-minute cache writes$1.25
Output$5.00

GPT-6.1 Terra or GPT-6.1 Sol?

Terra is a useful starting point for a bounded screen task, visual interpretation or a small writing change. OpenAI recommends Sol or Astra for the most complex coding work. Keep the task, tools, effort and acceptance test fixed when comparing models.

Sol Standard costs $2 per million fresh input tokens and $10 per million output tokens below the long-input tier. At the same 50,000-input and 5,000-output budget, Sol costs $0.15 and Terra $0.075. Actual task savings depend on how many turns and repairs each model needs. A cheaper failed attempt also adds time before escalation.

Context, media and reliability

The API window is 1,050,000 tokens, with a maximum of 922,000 input tokens and 128,000 output tokens. Image input is supported; direct audio and video input are not. Video editing through an application uses tools and screen observations.

In a launch-day Reddit report, Hefestion89 described a chapter-by-chapter novel review in ChatGPT Work that repeatedly ignored a request to avoid intermediate updates, even after changing from medium to maximum effort. This is one writing workflow, with no controlled computer-use completion rate. It is a reason to check procedural requirements as well as the final artifact.

Before you use it

  • The application must supply and operate the computer environment.
  • Plan allowances, model tokens and execution services have separate cost rules.

How to get started

Try one bounded task at medium effort, verify its saved result, and record all turns and elapsed time before increasing the workload.

Sources

Sources (5)

Calculators