OpenAI releases GPT-6.1 Terra at half Sol’s price. When to use it in Codex

What OpenAI released

On October 26, 2026, OpenAI introduced GPT-6.1 Terra, “a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing.” Its GPT-6 guide now lists four models: Astra for the most demanding work, Sol to balance “speed, cost, and intelligence,” Terra as “fast and efficient,” and Luna as the fastest and cheapest.

The Codex changelog says it “is now available in ChatGPT Work and Codex on Plus, Pro, Business, Enterprise, and Edu plans.” It “isn’t available in Chat,” and on Enterprise, access “depends on workspace availability and administrator controls.” In the API, it’s gpt-6.1-terra.

The name leaked first. On October 20, TestingCatalog reported that it had turned up in Microsoft’s evaluation settings and, citing a paywalled report in The Information, wrote that the new Terra was “expected to beat GPT-6.1 Sol while costing less.”

What it costs

Model	API, in / out per million tokens	Codex credits, in / out per million tokens	Plus, local Codex messages per five hours
GPT-6 Astra	$10 / $50	250 / 1,250	5–45
GPT-6.1 Sol	$2 / $10	50 / 250	15–160
GPT-6.1 Terra	$1 / $5	25 / 125	35–285
GPT-6 Luna	$0.10 / $0.50	2.5 / 12.5	350–3,000

The API prices are from OpenAI’s pricing page, and the Codex columns from Codex pricing. The message counts are OpenAI’s estimates for Plus and Standard Business, “not fixed message limits.” Pro plans currently have no five-hour limit, but their weekly allowance still applies.

Terra doesn’t add usage to a plan. “Overall Work and Codex plan limits are unchanged,” says the pricing page. It uses less of the allowance per task: in OpenAI’s words, “the same usage allowance covers more tasks than with GPT‑5.6 Terra and substantially more than with GPT‑6.1 Sol or GPT‑6 Astra.”

Against the Terra it follows, the new one is cheaper too. GPT-5.6 Terra costs $2 in and $12 out in the API, and 50 and 300 credits in Codex. Two API details from the model page: cached input costs $0.05 per million tokens, and prompts over 272,000 input tokens cost twice as much for input and 1.5 times as much for output. Fast mode works with Terra and uses included Codex usage at 2.5 times the Standard rate. Ultrafast is only for Astra and Sol.

How good it is

Every score here is OpenAI’s own, from its announcement.

• Coding. On Terminal-Bench 4.0 at maximum reasoning effort, Terra solves 50.3% of tasks, against 21.5% for GPT-5.6 Terra. OpenAI adds: “For the most complex coding tasks, GPT-6.1 Sol and GPT-6 Astra remain the better choice.”
• Computer use. On Agents’ Last Exam at medium effort, Terra scores 15 points higher than GPT-5.6 Terra at about a third of the cost, and “performs on par with GPT-6 Astra at only one-tenth its price per token.” At maximum effort, OpenAI puts its cost per task about nine times below Claude Opus 5.5’s, and says it finishes tasks more than three times faster than Astra and Sol.
• Science. On Terminal-Bench-Science 0.1 at maximum effort, Terra “surpasses GPT-6.1 Sol at lower cost,” within five points of Opus 5.5.
• Factual errors. On OpenAI’s internal set of error-inducing prompts, Terra’s error rate at low effort is 16.7%, down from 53.6% for GPT-5.6 Terra, and 4.3% at maximum effort.

So the report was half right. Terra beats Sol on OpenAI’s science test at half the price, but OpenAI itself still recommends Sol and Astra for the hardest code.

Where you can use it

• The ChatGPT desktop app. In Work or Codex, open Advanced and choose GPT-6.1 Terra from the model list.
• ChatGPT Work on the web and on mobile.
• The Codex CLI, version 0.159.0 or later. OpenAI’s models page gives the command: codex -m gpt-6.1-terra. In a running session, use /model. To make it your default, set model = "gpt-6.1-terra" in config.toml, which the desktop app, CLI and IDE extension share.
• Not the Codex IDE extension, for now. The models page lists Terra for the desktop app, Work on the web, the CLI, credits and the API, but not for the IDE extension, which it lists for Astra, Sol and Luna.
• The API, as gpt-6.1-terra, with a 1.05-million-token context window, 128,000 output tokens and a knowledge cutoff of June 1, 2026. Reasoning effort runs from none to max, with medium the default, and tool calling needs the Responses API. Codex signed in with an API key can use it too, since there “model availability follows your API organization and project.”
• GitHub Copilot, on Copilot Pro, Pro+, Max, Business and Enterprise, in VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, the Copilot CLI, the cloud agent, github.com and GitHub Mobile. GitHub added it on October 26 and says “rollout will be gradual.” It bills Terra at the provider’s list price under Copilot’s usage-based billing, and on Business and Enterprise it’s on unless an administrator has turned off new models or this one.
• Not in Chat, on Free or Go, or in Cursor. Codex pricing lists GPT-6 Luna for Free and Go instead, in the desktop app and “subject to rollout.” OpenAI stopped giving Cursor new models when it planned its cutoff, and the GPT models in Cursor’s price table are still GPT-5.6 Sol, Terra and Luna.

When to pick Terra, and when Sol

OpenAI’s models page sorts them by the work. Astra is “for the hardest end-to-end work.” Sol is “for repeated, long-running work” across code, apps and documents “when cost matters.” Terra is “for everyday writing and visual work”: “interpreting images and screens, and completing tasks across supported apps.” Luna is for “clear, repeatable tasks.”

In practice:

1. Screenshots, charts, forms, drafts and work in other apps: Terra. That’s what OpenAI built it for, and it costs half as much as Sol.
2. Routine code, such as a small fix or a test: try Terra on a task you know well. Its coding score more than doubled over GPT-5.6 Terra’s.
3. Hard, multi-file changes and stubborn bugs: Sol or Astra, as OpenAI recommends.
4. High-volume, well-defined jobs like extraction or classification: Luna.

On Plus, which has a five-hour limit, moving routine work to Terra leaves more of each window for Sol. OpenAI suggests starting at medium effort and raising it “if the improvement in quality is worth the added time and tokens.”

How careful it is

OpenAI says Terra “meets the Critical threshold in cybersecurity under our Preparedness Framework,” and uses the same safeguards as Sol and Astra. In the tests in its system card, Terra made no attempts to get around Codex’s auto-review after it denied a command.

It was less careful with barriers in the environment, such as an access-denied error. “In 29.6% of rollouts at maximum reasoning effort, GPT-6.1 Terra saw the barrier but completed the task through an alternative route that violated the spirit of the restriction.” That’s more often than Astra, GPT-6 Luna and GPT-6.1 Sol, and less often than GPT-6 Sol. OpenAI notes the test covers low-stakes situations and runs without the system-level controls that would “plausibly stop” many of those workarounds. If you give Terra computer use, keep the same sandbox and approvals you’d give any agent.

Trying Terra beside Sol

Enjoy runs Codex on your ChatGPT plan, and its model list comes from Codex itself, so Terra appears there when your Codex offers it, without an Enjoy update. If it’s missing, update Codex from the agent settings: Terra needs Codex 0.159.0 or later.

A conversation can use a different model for each message, so Terra can read the screenshots and Sol can make the hard fix in the same thread. When Codex reaches its limit, the conversation waits for the reset and carries on by itself, as long as the computer is awake and Enjoy is open, or you can switch it to another agent, such as Claude Code. Enjoy is free, and it doesn’t add usage to a ChatGPT plan.

What OpenAI hasn’t said

• Why the announcement leaves out Business. The changelog and Codex pricing page list Business plans, but the announcement names only “Plus, Pro, Enterprise, and Edu users.”
• When it reaches the Codex IDE extension, if it will.
• What happens to GPT-5.6 Terra. OpenAI hasn’t announced a retirement, and the Codex pricing page says its rates “remain unchanged.”
• Whether Terra gets Ultrafast. For now, Ultrafast covers only Astra and Sol.

We’ll update this memo when OpenAI fills these in.