GPT-6.1 Terra

GPT-6.1 Terra is OpenAI’s October 26, 2026 model for clear writing, visual reasoning and computer use. Its exact API ID is gpt-6.1-terra. Use Terra in ChatGPT Work, Codex or the API; ordinary Chat is a separate product surface. Start here for concrete settings and costs, or compare the series in the GPT-6.1 overview.

GPT-6.1 Terra pricing

Standard API prices in USD per million tokens, before long-context and regional adjustments:

Token category	Price
Ordinary input	$1.00
Cached input	$0.05
Cache writes, 30-minute minimum lifetime	$1.25
Cache writes, 12-hour minimum lifetime	$2.50
Output, including billable reasoning	$5.00

A request with 10,000 ordinary input tokens and 2,000 output tokens costs $0.02 in these text-token charges. At 1,000 equivalent requests, that is $20. This arithmetic assumes Standard processing and no cache use; image processing and tools can add charges. Use the pricing calculator with your actual input and output counts.

When input exceeds 272,000 tokens, the full request uses twice the input and cache rates and 1.5 times the output rate. Exactly 272,000 input tokens stays at the short-context rate. Batch and Flex rates are half Standard, Fast rates are twice Standard, and eligible regional processing adds 10%. A large context limit does not mean every request has the same per-token cost.

The official model reference records a 1,050,000-token combined context window and a 128,000-token maximum output. Reserved output shares the context window with input. Plan enough space for the response and tool results when sending a long document collection.

Access Terra in Work and Codex

Terra is available on Plus, Pro, Business, Enterprise and Edu in Work and Codex. In the desktop app or Work on the web, open Advanced and choose GPT-6.1 Terra. Work also supports Terra on mobile. Enterprise availability depends on workspace access and administrator controls. Terra does not support UAE inference residency.

For Codex CLI, use codex -m gpt-6.1-terra or choose the model with /model in an interactive session. CLI 0.166.0 includes Terra in the bundled, Vertex AI and Amazon Bedrock catalogs. A catalog listing does not grant provider or workspace access. Confirm the active account, provider and permissions if the model is missing.

Standard Codex credit rates are 25 input, 1.25 cached-input and 125 output credits per million tokens. There is no separate Codex cache-write charge. API-key billing follows the dollar rate card instead. The pricing guide explains the subscription allowance, paid-credit and API budgets.

Make a first API request

Use the Responses API when the workflow needs tools. Tool calling requires Responses; a model’s ability to understand a screenshot does not grant control of a computer by itself. Configure the actual tools and the application’s action permissions before evaluating an agent workflow.

import OpenAI from "openai";
const client = new OpenAI();
const response = await client.responses.create({
  model: "gpt-6.1-terra",
  reasoning: { effort: "medium" },
  input: "Rewrite this update for a customer: The delivery moved to Friday."
});
console.log(response.output_text);

This is an illustrative request, not a recorded benchmark. The model supports reasoning efforts none, low, medium, high, xhigh and max, with medium the API default. Start with the default and increase effort when your evaluation shows that a harder task benefits from it. For credentials, organization selection and common errors, follow the API setup guide.

Terra versus Sol: evaluate the work you repeat

Terra’s Standard input and output rates are half GPT-6.1 Sol’s. That makes Terra worth evaluating for recurring writing, visual interpretation and computer tasks. Sol remains another option for complex coding and sustained professional workflows. The useful comparison is the total cost of a correct result, including retries and review time.

For a writing task, save the same brief, audience and source facts. Compare whether both models retain names, dates and required actions. For a visual task, supply the same screenshot and compare extracted labels, values and uncertain details against the image. For an app task, define the permitted actions and inspect the resulting document or application state. A polished response alone does not prove that the intended change happened.

Keep the successful prompt and validation checks with the workflow. Re-run a small representative sample before changing a production default, especially when the workflow edits files, enters records or takes actions in another application.