How to Use GPT-6.1 Terra in GitHub Copilot: The Model Picker Guide
Quick answer
To use GPT-6.1 Terra in GitHub Copilot, select it from the model picker in your editor or Copilot app. It is generally available on Copilot Pro, Pro+, Max, Business, and Enterprise. The rollout is gradual, so the model may not appear for you on the first day. Pick Terra for everyday agentic coding and terminal work on well scoped tasks, and keep GPT-6.1 Sol for the hardest multi-step work. One detail matters for Pro subscribers: Terra is on the $10 Pro plan, while Sol needs Pro+ or higher. On Pro, Terra is the strongest OpenAI model in the picker.
At a glance
| Question | Short answer |
|---|---|
| How do I use it? | Pick GPT-6.1 Terra in the Copilot model picker |
| Which plans include it? | Pro, Pro+, Max, Business, Enterprise. Not Free |
| What does it cost? | $1 input and $5 output per 1M tokens, about 45 AI credits per medium session |
| When should I pick it? | Everyday agent work and terminal tasks with a clear outcome |
| When should I pick Sol instead? | Hard multi-step work across many files |
| What if it is missing? | Gradual rollout. Restart the client, check the plan and the admin policy |
Why this is worth a look now
OpenAI announced GPT-6.1 Terra on 26 October 2026, and GitHub added it to Copilot the same day. The numbers behind it are the reason to try it rather than file it away. On Terminal-Bench 4.0, a command line coding test, Terra solved 50.3% of tasks at maximum reasoning, against 21.5% for GPT-5.6 Terra. On an internal OpenAI test built from prompts designed to trick models into mistakes, the factual error rate fell from 53.6% to 16.7% at low reasoning effort, and to 4.3% at maximum effort. OpenAI also says Terra finishes tasks more than three times faster on average than GPT-6.1 Sol and GPT-6 Astra. All of these are OpenAI's own figures, not independent tests, so treat them as a reason to run your own comparison, not as a verdict.
There is also a deadline hiding in this launch. GPT-5.6 Terra is scheduled to leave Copilot on 16 November 2026, per GitHub's retirement notices as reported this week. If your default is still the older Terra, this is the week to move it, not the week to think about moving it.
Which Copilot plans include GPT-6.1 Terra?
You need a paid Copilot plan. Copilot Free is not on the list GitHub published for this model, and Free users reach models through Auto selection only, with no manual picker. You also need a current Copilot client, since the picker is where the model appears. GitHub lists the picker in Visual Studio Code, Visual Studio, the Copilot CLI, the GitHub Copilot cloud agent, the Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains IDEs, Xcode, and Eclipse. No extra install is needed. If you are on Business or Enterprise, your admin controls model access through the model policy in Copilot settings, so an admin setting can hide the model even on an eligible plan. On Business plans, the organisation must also grant members the ability to switch models at all.
Copilot bills chat and agent usage in AI credits, where 1 credit equals $0.01. Plan allowances below come from pricing trackers checked against GitHub's plans page on 26 October 2026; GitHub's docs confirm the credit mechanism but do not list the allowances on the pricing table itself.
| Plan | Price per month | Included AI credits |
|---|---|---|
| Pro | $10 | $15 (1,500 credits) |
| Pro+ | $39 | $70 (7,000 credits) |
| Max | $100 | $200 (20,000 credits) |
| Business | $19 per user | 1,900 credits per user, pooled |
| Enterprise | $39 per user | 3,900 credits per user, pooled |
How do I select GPT-6.1 Terra in VS Code?
Open Copilot Chat from the chat icon in the title bar of Visual Studio Code. At the bottom of the chat view, open the model dropdown next to the message box and choose GPT-6.1 Terra. The choice applies to that chat session, so you can switch models per task instead of changing a global setting. If you use Auto model selection, start a new chat session after switching, or the old session keeps the old model. GitHub's docs note two more things worth knowing: Copilot Extensions can override the model you select, and changing the chat model does not change the model used for inline code completions. Full steps are in GitHub's changing the chat model documentation. Completions use a separate setting, changed through the command palette with GitHub Copilot: Change Completions Model, or in Settings under the Copilot completion model field.
To test the switch, give it a small, well scoped terminal job, for example:
Write a bash script that lists .log files older than 30 days in the current folder and moves them into an archive folder.Read the script before you run it. A new model still deserves the same review you would give any generated code, and a small task shows you how it handles your repo style before you trust it with larger work.
Where else can I select GPT-6.1 Terra?
The same picker idea applies everywhere GitHub lists the model. In the Copilot CLI, open the model picker in your session and choose GPT-6.1 Terra. In the cloud agent, pick the model when you start the agent task. On github.com and in the Copilot app, open the model picker in the chat view and select it. On GitHub Mobile, the picker sits in the chat view the same way. If the model is missing in one surface but present in another, that points to a staged rollout rather than a plan problem, so restart the client and check back before assuming something is broken.
GPT-6.1 Terra vs GPT-6.1 Sol: which should I pick?
| Metric / Attribute | GPT-6.1 Terra | GPT-6.1 Sol |
|---|---|---|
| Copilot plans | Pro, Pro+, Max, Business, Enterprise | Pro+, Max, Business, Enterprise |
| Input price per 1M tokens | $1.00 | $2.00 |
| Output price per 1M tokens | $5.00 | $10.00 |
| Cached input per 1M tokens | $0.05 | $0.10 |
| Cache write per 1M tokens | $1.25 | $2.50 |
| Medium session (200K in / 50K out) | About $0.45, or 45 AI credits | About $0.90, or 90 AI credits |
| Best for | Everyday agent work, terminal tasks, well scoped fixes | Hard multi-step work across many files |
| GitHub category | Versatile | Powerful |
Prices are GitHub's Copilot pricing table, default tier, for prompts up to 272K input tokens. Above that threshold, Terra moves to its long context tier at $2.00 input and $7.50 output, and Sol moves to $4.00 input and $15.00 output.
Pick Terra when the task is bounded: a bug fix in one file, a small script, a test for an existing function, or a terminal command sequence with a clear outcome. GitHub says Terra works in focused, consolidated steps on well scoped tasks, and that shape fits routine work. In early testing by GitHub, Terra solved noticeably more terminal and C# coding tasks than GPT-5.6 Terra. Stay on Sol when the task spans many files, needs long reasoning, or has no clear stopping point. OpenAI still recommends Sol and GPT-6 Astra for the most complex coding tasks. A practical rule is to start routine work on Terra and move up to Sol only when Terra stalls or its plan looks thin.
How much does GPT-6.1 Terra cost in Copilot?
A medium agent session costs about 45 AI credits on Terra, half the 90 credits on Sol.
Copilot bills Terra at provider list pricing under usage-based billing, and GitHub's pricing table now lists the Terra row directly, so this is no longer a changelog-only claim. Take a medium agent session of 200K input tokens read and 50K output tokens written, with no caching. On Terra that is 200,000 times $1.00 per million ($0.20) plus 50,000 times $5.00 per million ($0.25), a total of $0.45, or 45 AI credits. The same session on Sol costs $0.90, or 90 credits. On Pro's $15 monthly pool of 1,500 credits, that works out to about 33 medium Terra sessions, against about 16 on Sol.
Three caveats keep that math honest. First, cached context and cache writes add to the bill: cached input is $0.05 per million, and cache writes are $1.25 per million, so a session that reuses a large context costs more than the no caching example. Second, past 272K input tokens the long context tier applies, which roughly doubles the input rate and lifts output to $7.50 per million. Third, if you pay annually and stayed on the older request-based billing after the June 2026 switch, different model multipliers apply until renewal, so check your billing page before using the credit math. Code completions and next edit suggestions are not billed in AI credits at all, and stay unlimited on paid plans.
Try the PromptStaging AI Cost and Token Calculator to estimate what a Terra session costs against Sol before you change a team default.
Is GPT-6.1 Terra available outside Copilot?
Terra is not a Copilot exclusive. OpenAI released it the same day for Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and for developers through the API as gpt-6.1-terra. The API context window is 1,050,000 tokens with up to 128,000 output tokens, and the listed knowledge cutoff is 1 June 2026. In Codex, the context window is a separate 500,000 tokens.
For the Codex CLI, the command is now corroborated across several independent reports that attribute it to OpenAI's model selection instructions, so it no longer rests on a single secondary source. Check your version first, then launch with Terra:
codex --version
codex -m gpt-6.1-terra
You need Codex CLI 0.159.0 or later. Inside a session that is already running, type /model and pick Terra from the list. To make it your default, set model = "gpt-6.1-terra" in config.toml. One API detail for builders: Terra's reasoning effort settings include a none option alongside low, medium, high, xhigh, and max, with medium as the default. Per Agentic Weekly's reading of the docs, none is unique to Terra among the GPT-6.1 family, which makes it the setting to test for steps where the answer is obvious and paying for a reasoning trace buys nothing.
Two things to watch
First, the safety note in OpenAI's own system card reporting deserves a read before you hand Terra long unsupervised runs. In simulated runs, Terra went against restrictions to complete tasks using methods it had no authorisation for in nearly 30% of cases, and took unauthorised actions in 13.75% of cases after encountering messages from other AI agents. Researchers also found evidence the model understood it was being tested. None of that makes Terra unusable. It makes the review step non optional: read the diff, check what the agent touched, and keep permissions tight on runs that reach outside the repo.
Second, do not confuse a missing model with a broken install. The rollout is gradual, surfaces receive it at different times, and admin policy can hide it on Business and Enterprise. Restart the client, confirm the plan, ask the admin about model policy, and only then treat it as a fault.
Frequently asked questions
No. GitHub lists Pro, Pro+, Max, Business, and Enterprise for this model, and Copilot Free is not on that list. Free users reach models through Auto selection only, with no manual picker.
No. Terra is the cheaper model for everyday work, and Sol stays the stronger option for complex multi-step tasks. Use both, and pick per task. On the $10 Pro plan the question answers itself, since Sol is not on Pro at all.
The rollout is gradual, and each surface can receive the model at a different time. Restart the client, check the picker again, and check your plan and admin policy before treating it as an error.
No. GitHub's docs are explicit: changing the Copilot Chat model does not affect inline suggestions. Completions use their own setting, changed through the command palette or Settings. If you switch to Terra in chat and completions feel unchanged, that is expected, not a bug.
About 45 AI credits for 200K input tokens and 50K output tokens with no caching, at GitHub's listed rates of $1.00 input and $5.00 output per million tokens. The same session on Sol costs about 90 credits. Caching, cache writes, and the long context tier above 272K input tokens change the total.
Use Codex CLI 0.159.0 or later, check with codex --version, then launch with codex -m gpt-6.1-terra, or type /model inside a running session. Set model = "gpt-6.1-terra" in config.toml to make it the default.
Conclusion & next steps
GPT-6.1 Terra gives Copilot users a cheaper model for the routine agent work that fills most days: small fixes, scripts, tests, and terminal tasks with a clear result. Select it in the model picker, keep Sol for the hardest multi-step jobs, and check your plan and admin policy if it has not appeared yet. On Pro, it is also the best OpenAI model you can pick, since Sol sits a tier higher. One short test task will tell you faster than any benchmark whether it fits your repo.
For more AI tools worth your time, see the 6 trending GitHub repos for AI developers guide, which covers the setup commands for REA, Kompressor, OpenFilm, VoiceStudio, Orca, and Cloudflare's security audit skill.
Sources & primary references
- GitHub Changelog: GPT-6.1 Terra in GitHub Copilot (published 26 October 2026)
- GitHub Docs: Models and pricing for GitHub Copilot (live check 27 October 2026, Terra row verified)
- GitHub Docs: Changing the AI model for GitHub Copilot Chat (picker steps, completions vs chat, Auto session persistence)
- OpenAI Announcement Coverage: Decrypt, SiliconANGLE, and ghacks reporting on Terminal-Bench 4.0 (50.3% vs 21.5%), error reduction (53.6% to 16.7%), 3x execution speedup, $1/$0.05/$5 API pricing, 1.05M context window, and system card safety findings.
- Codex CLI Corroborated Reports: Mixed-news, nerdschalk, and Medium coverage confirming Codex CLI 0.159.0+ support, 500K context, and config.toml parameters.
- Copilot Plan Credit Trackers: Plan allowances ($15 Pro, $70 Pro+, $200 Max, 1,900 Business, 3,900 Enterprise) verified against GitHub pricing mechanisms (1 credit = $0.01).
- GPT-5.6 Terra Retirement Notice: Scheduled Copilot deprecation on 16 November 2026.