GPT-6.1 Terra is out: half the API price of Sol, built for computer use
OpenAI released GPT-6.1 Terra on October 26, the cheaper model in the GPT-6.1 family. It is aimed at everyday work on a computer: filling forms, reading drawings, writing proposals. In the API it costs half of what GPT-6.1 Sol does, and it is already in ChatGPT Work, Codex and GitHub Copilot.
What happened
OpenAI published GPT-6.1 Terra on October 26, 2026, and made it available the same day. Terra is the fast, low-cost member of the GPT-6.1 family, next to the larger GPT-6.1 Sol. It replaces GPT-5.6 Terra, and OpenAI pitches it at ordinary work on a computer: drafting proposals, reading technical drawings, filling in forms and editing video.
The headline is cost. OpenAI says the standard API price is 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol, and that inside ChatGPT Work and Codex the same usage limit stretches to more tasks than with GPT-5.6 Terra, and many more than with GPT-6.1 Sol or GPT-6 Astra.
Where you can use it
- ChatGPT Work and Codex: Plus, Pro, Enterprise and Edu plans, from launch day.
- API: the model name is
gpt-6.1-terra. - GitHub Copilot: generally available from October 26 on Pro, Pro+, Max, Business and Enterprise, rolling out gradually. It shows in the model picker in VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI, the cloud agent, github.com and GitHub Mobile. On Business and Enterprise it switches on by default unless an admin has turned off new models.
| API, per 1M tokens | Input | Cached input | Output |
|---|---|---|---|
| gpt-6.1-terra | $1.00 | $0.05 | $5.00 |
The numbers OpenAI gives
All from OpenAI's launch post. The fact-error test is internal; the rest are public benchmarks run by OpenAI.
| Test | GPT-5.6 Terra | GPT-6.1 Terra | What it says |
|---|---|---|---|
| Terminal-Bench 4.0, max reasoning | 21.5% | 50.3% | More than double on coding in a terminal. Sol and Astra are still OpenAI's pick for the hardest coding. |
| Factual errors on trap prompts, low reasoning (internal, lower is better) | 53.6% | 16.7% | At max reasoning the rate is 4.3%, close to Astra and Sol. |
| Agents' Last Exam, medium reasoning | baseline | +15 points | At about a third of the cost; level with GPT-6 Astra at a tenth of the price per token. |
OpenAI also compares Terra with Anthropic's model directly: at maximum reasoning on Agents' Last Exam, it says Terra uses tokens more than three times as efficiently as Claude Opus 5.5, at a quarter of the price per token, so a task costs about nine times less. On Terminal-Bench-Science it puts Terra within five points of Opus 5.5 at more than five times lower estimated cost. Those are OpenAI's own runs, so read them as a claim to check on your work, not a verdict.
One safety note from the post: Terra reaches the “critical” cybersecurity capability threshold in OpenAI's Preparedness Framework and ships with the safeguards built for GPT-6 Astra.
What it means for you
- Paying for GPT-6.1 Sol in the API for routine jobs? Try Terra on a sample of real tasks. At half the price it is the obvious first test for bulk work like reading invoices, filling forms or tagging documents.
- On ChatGPT Plus and hitting the Codex limit? Pick Terra for everyday tasks and keep Sol for hard bugs and refactors. The same quota goes further.
- On Copilot Business or Enterprise? Check the model policy in Copilot settings: under default enablement Terra is already switched on for your team.
- Choosing between Codex and Claude Code? The price gap just got wider on OpenAI's side. Our Claude Code vs Codex comparison has the plans, limits and where each one runs.
The Codex review lists what each ChatGPT plan includes. For prompts that keep a coding agent on a short leash, see the Claude Code prompts; they work in Codex as written.
Try it
Below is the text of a form, invoice or document. Extract every field into a table with three columns: field name, value, and how sure you are (high / medium / low). Rules: - Copy values exactly as written. Do not fix spelling, round numbers or convert currencies. - If a field is missing or unreadable, write MISSING. Do not guess. - After the table, list anything that looks inconsistent (totals that don't add up, dates in the wrong order, two different names for the same party). Document: [paste the text here]
I'm going to give you a list of tasks from my week. For each task, decide whether a fast, cheap model can do it well or whether it needs the strongest model I have. Answer as a table: task, "cheap" or "strong", and one line on why. Mark as "strong" anything with: multi-step reasoning across files, code that touches money or security, or a decision I can't easily check. Mark as "cheap" anything that is extraction, formatting, summarising or a first draft I will edit. Tasks: [list your tasks, one per line]
Sources
- OpenAI, Introducing GPT-6.1 Terra, October 26, 2026
- GitHub Changelog, GPT-6.1 Terra in GitHub Copilot, October 26, 2026