How to Use GPT-6.1 Terra for Agentic Coding Tasks

OpenAI released GPT-6.1 Terra on 26 October 2026 as a fast mid-tier model built for visual reasoning and computer use. It replaces GPT-5.6 Terra and costs half as much on the API while finishing many desktop and coding tasks faster than higher-end models.

This guide shows how to access it, set it up for agentic workflows, write effective prompts, and keep costs under control.

What GPT-6.1 Terra Offers

GPT-6.1 Terra combines detailed screen understanding with the ability to click, type, and navigate apps. OpenAI positions it for routine professional work: drafting documents, interpreting drawings, filling forms, and editing media.

Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. The model ID is gpt-6.1-terra. Context window reaches 1,050,000 tokens with up to 128,000 output tokens. Knowledge cutoff is 1 June 2026.

On Agents’ Last Exam, it scores 15 points higher than GPT-5.6 Terra at medium reasoning effort and roughly one-third the cost. At maximum effort it finishes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol on OpenAI’s computer-use evaluations. Terminal-Bench 4.0 scores 50.3% at max effort, more than double the previous Terra result.

How to Access GPT-6.1 Terra

Paid ChatGPT users get it immediately in ChatGPT Work and Codex. That covers Plus, Pro, Enterprise, and Edu plans. Free and Go users do not have access in the standard chat interface.

Developers use the OpenAI API with the model string gpt-6.1-terra. It supports the Responses API, which exposes computer use, web search, code interpreter, hosted shell, and MCP tools.

In Codex, switch the model selector to GPT-6.1 Terra when you start a new task or continue an existing session. In ChatGPT Work, select it from the model menu before assigning a multi-step job.

Setting Up for Agentic Coding

Agentic coding means giving the model a clear goal, the tools it needs, and a finish condition, then letting it run. Terra works well for this because computer use is built in and the cost stays low for repeated steps.

Start with a short, concrete task. Example: “Read the error log in the terminal, identify the failing test, fix the assertion, and run the test suite. Stop when all tests pass.”

Provide the relevant files or a repository path. Set reasoning effort to medium for most work; raise it to high or max only when the task involves dense visual analysis or long debugging chains.

Use prompt caching for any repeated context such as system instructions or large codebases. Cached input drops to $0.05 per million tokens, which keeps multi-turn agent loops affordable.

Writing Effective Prompts

Give the model a clear finish line. State what “done” looks like and what to do if it hits a blocker. Terra is better at staying in scope and reporting limitations than earlier Terra versions.

Example prompt for a coding agent:

You are a coding agent. Goal: add input validation to the user registration endpoint.
Constraints: keep the existing API response format, add unit tests, do not change unrelated files.
Finish when the new tests pass and the endpoint rejects invalid emails and empty passwords.
If you need a file that is not open, request it by path.

For computer-use tasks, describe the screen state you expect and the next action. “The browser is open to the settings page. Click the Security tab, then enable two-factor authentication for the admin account. Confirm the success message appears.”

Break long jobs into bounded subtasks. Review the output of each step before letting the agent continue. This catches errors early and limits token waste.

API Usage Pattern

Here is a minimal Python example using the Responses API for a simple agent step:

from openai import OpenAI
client = OpenAI()

response = client.responses.create(
    model="gpt-6.1-terra",
    input="Inspect the open terminal and list any failing tests.",
    tools=[{"type": "computer_use"}],
    reasoning={"effort": "medium"}
)
print(response.output_text)

Add tool results back into the next request so the model can continue. For high-volume runs, use the Batch API for a 50% discount on both input and output.

Prompts longer than 272,000 input tokens are billed at double the input rate and 1.5× the output rate for the entire request. Keep contexts under that threshold when possible, or rely on caching.

Verification and Cost Control

Always verify agent output. Ask Terra to show the exact command it ran and the resulting output. Cross-check critical changes yourself, especially anything that touches production data or security settings.

Track cost by logging input and output tokens. At $1/$5 rates, a typical agent loop of 10k input and 2k output tokens costs about $0.02. Cached prefixes drop that further. Compare this with higher-tier models when the task does not need maximum capability.

For the hardest coding problems, OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra. Terra covers the high-volume middle ground where speed and price matter more than the last few points of accuracy.

If you already work with other models for coding, the patterns transfer. See our earlier guide on using Claude Opus 5.5 for agentic coding for complementary prompting techniques.

Tips for Better Results

• Start every session with a short system prompt that defines the role, tools, and stop conditions.
• Prefer medium reasoning effort unless the task involves complex visual layout or scientific calculation.
• Use structured outputs or function calling when you need reliable machine-readable results.
• Monitor for prompt injection; Terra shows strong resistance in OpenAI’s tests, but still review unexpected actions.
• Rotate API keys and limit computer-use permissions to the apps the agent actually needs.

Conclusion

GPT-6.1 Terra gives developers a practical option for agentic coding and computer-use tasks at a price that supports high-volume work. Access it through ChatGPT Work, Codex, or the API as gpt-6.1-terra. Write prompts with clear goals, use caching, and verify each step. The result is faster progress on everyday engineering work without the cost of the top-tier models.

Sources

• Introducing GPT-6.1 Terra
• Addendum to GPT-6 Astra System Card: GPT-6.1 Terra
• OpenAI API documentation