Michael JohnsonChoosing a model in a coding assistant affects two practical factors at once: how well it completes a task and how much each use costs. On October 26, 2026, GitHub announced the general availability of GPT-6.1 Terra in Copilot and described it as suited to everyday agentic coding and terminal work. For teams, the practical question is now specific: which tasks can Terra handle well enough to become the default choice?
What GitHub announced
According to GitHub’s October 26 announcement, GPT-6.1 Terra is available to Copilot Pro, Pro+, Max, Business, and Enterprise subscribers. It can be selected in VS Code, Visual Studio, Copilot CLI, the Copilot cloud agent, and other GitHub clients listed in the announcement. The rollout is gradual, so users may not get access at the same time.
GitHub positions the model for everyday agentic work and terminal tasks, noting that it costs less than GPT-6.1 Sol. The company also says that in early testing, Terra solved noticeably more terminal and C# tasks than GPT-5.6 Terra. This is GitHub’s claim about early tests: the published notice does not provide the size of the test set, metrics, or comparison conditions. It is therefore best treated as a cue for choosing tasks for your own pilot, with results measured in a working repository.
Access and billing
GitHub says that billing follows the provider’s rates on a usage-based basis. The specific cost depends on the current terms and how Copilot bills for model usage; before running a pilot, check GitHub’s model documentation and current Copilot pricing.
In Business and Enterprise organizations, administrators manage access through model policies in Copilot settings. When the global default is enabled, the new model is turned on automatically unless an administrator disables that rule or specifically blocks Terra. So if the model does not appear in the list, check both the user’s access and the organization’s policy, and allow for the gradual rollout.
How to run a short pilot
Choose a set of real, well-scoped tasks: fix a small bug through the terminal, change C# code, update a test, and handle a task that requires the agent to explore the repository before making an edit. This set reflects the scenarios GitHub highlights and helps evaluate the model against your codebase.
- Define the task and success criteria. For each example, record the expected change, required tests, and the conditions that count as completion.
- Compare models under the same conditions. Use the same prompt, branch or repository snapshot, and tool permissions. Save the diff and the agent’s activity log.
- Evaluate the result, not the impression. Measure the share of tasks that pass tests and review without substantial rework. Track hands-on time, the number of retries, and cases where the agent chose the wrong terminal action separately.
- Weigh quality against cost. Use actual billing data from the pilot period and calculate the cost per accepted task. This gives the team a practical comparison instead of relying only on the model’s price or the speed of individual responses.
If Terra delivers an acceptable level of fixes at lower cost on typical tasks, make it the default for those scenarios while keeping a more capable model for complex work. If the benefit does not show up in your own code, let developers choose their model and repeat the evaluation with a different task set.
What to take away from the update
GPT-6.1 Terra is now available in GitHub Copilot for the listed plans and client applications, with everyday agentic coding and terminal work as its stated focus. Start by checking your organization’s policy and running a small, like-for-like pilot: it will show how well the model’s stated positioning matches your team’s tasks and the actual cost of an accepted result.
