AI, in your work.

What GitHub announced

GitHub announced general availability on October 26 of GPT-6.1 Terra in Copilot for agentic coding and terminal work.

The company says its early tests achieved better results than GPT-5.6 Terra on terminal and C# tasks, without publishing figures in that announcement. This provides a reason to evaluate the model, but does not establish which model suits each team.

Availability and access policies

The announcement includes Copilot Pro, Pro+, Max, Business and Enterprise. Announced clients include Visual Studio Code, Visual Studio and Copilot CLI. Rollout is gradual.

There is a relevant exception: the documentation for legacy annual plans says subscribers retaining request-based billing do not receive new models or features. The plan name alone does not confirm access.

For Business and Enterprise, administrators can enable or restrict models. The compatibility documentation explains that unconfigured generally available models inherit the default policy, while explicit choices take precedence. The table checked does not yet specify minimum IDE versions for Terra.

Before the pilot, check the model picker in the team's intended client, its version and the effective model policy. These are environment-specific conditions that public sources cannot verify for an individual account.

Verified rates: four token categories

The Copilot pricing table, checked on October 27, lists these GPT-6.1 Terra rates in USD per million tokens. The tier depends on the number of input tokens: default, ≤ 272,000 tokens; long context, > 272,000 tokens.

Nexo's interpretation: comparing cost per task requires recording total consumption, including additional attempts and context sent. An individual rate cannot establish the cost of producing an acceptable change.

Included credits and spending limits

According to the organization billing documentation, Business and Enterprise pool included credits at the billing entity level. One credit equals USD 0.01.

Additional usage is enabled by default. When allowed, usage continues to be billed after the included pool is exhausted; when disabled, usage is blocked until the next billing cycle. Budgets can also stop access before the shared pool is exhausted.

For a limited evaluation, Nexo proposes setting a budget and recording both the calculated consumption cost and any additional charges. This distinguishes consumption covered by the plan from extra spending caused by the pilot.

A practical comparison proposed by Nexo

The following cases are synthetic examples designed to evaluate the model. They do not represent tests conducted or results obtained by Nexo.

Compare Terra with the team's usual model on the same project version, using identical instructions, tools and acceptance criteria. Start separate sessions and record the context and reasoning settings supported by each client.

• Define acceptance criteria before generating responses and retain tests the model cannot modify to approve its own solution.
• Repeat each case with both models and record every attempt, including failed runs.
• Record functional correctness, regressions, human review time, required interventions and cumulative cost per accepted task.
• Keep unresolved failures in the assessment: excluding them can make an alternative appear more efficient.

What a Business Central team can conclude

The announcement provides no AL-specific results. Nexo's interpretation is that Business Central teams should add their own AL cases before extending any conclusion to their extensions.

For example, they could evaluate validation using fictional records and check compilation, tests and expected behavior in their development environment. This example defines a possible evaluation; it does not establish model compatibility or improvements.

The proposed editorial decision criterion is specific: assess whether the model completes the selected tasks at a cost and with human review effort the team considers acceptable. A small comparison can justify a limited pilot; broader use will require tasks representative of real work.