GitHub announced on July 9, 2026 that OpenAI's GPT-5.6 family is rolling out in GitHub Copilot. The rollout includes Sol, Terra, and Luna, giving developers a way to choose between a higher reasoning ceiling, a balanced everyday model, and a lower-cost fast model.
GitHub describes the split directly. GPT-5.6 Sol has the highest reasoning ceiling and is best suited to large codebases, complex reasoning, and long-running agentic work. GPT-5.6 Terra is the balanced default for everyday interactive and agentic coding. GPT-5.6 Luna is the lightweight, cost-efficient option for smaller and faster tasks.
That split matters because Copilot is no longer only autocomplete. It now spans chat, CLI, cloud agent, app surfaces, github.com, IDEs, mobile, and review workflows. Each surface has different model needs. A large refactor may justify deeper reasoning. A terminal task needs reliable tool use. A mobile question may need speed and cost control more than maximum reasoning.
The availability list is also important: Visual Studio Code, Visual Studio, Copilot CLI, GitHub Copilot cloud agent, GitHub Copilot app, github.com, GitHub Mobile, JetBrains, Xcode, and Eclipse. Model choice is becoming a platform-wide Copilot capability rather than a feature attached to one editor.
Enterprise administrators should note the access policy. GitHub says Copilot Business and Enterprise admins must enable the GPT-5.6 model policy in Copilot settings, and that the policy is off by default. That is a sensible control point because new models affect cost, governance, developer access, and procurement expectations. The deeper agent workflows go into engineering, the more model switches need to be managed.
Usage Based Billing is the other practical detail. GitHub says these models are billed at provider list pricing. For engineering leaders, that means the model decision is no longer purely technical. It needs to match task value. Sol may be worth it for a difficult bug or a large architecture change. It is probably excessive for formatting, summarization, or low-risk batch work.
The bigger operational change is that coding agents now need model routing. Developers are not only asking whether Copilot has a given model. They are asking which level of model should handle this task. As model pickers appear across more development surfaces, routing will become part of normal engineering workflow design.
In the short term, this is a new model rollout for Copilot. Longer term, it shows coding-agent platforms moving toward finer resource allocation: high-reasoning models for complex work, balanced models for daily coding, and low-cost models for high-volume or low-risk work. That will shape the cost governance, speed, and reliability of AI development tools.
