GPT-5.6 in GitHub Copilot: Sol, Terra, or Luna?

GitHub added GPT-5.6 Sol, Terra, and Luna to Copilot's model picker, offering tiers from high-capability reasoning to low-cost fast tasks. The models carry different credit costs, with Sol requiring Pro+ plans and Terra serving as the everyday default. The rollout is gradual, with Enterprise admins needing to enable the policy.

GitHub added a changelog entry today that most developers will skip past: GPT-5.6 Sol, Terra, and Luna are now in the Copilot model picker https://github.blog/changelog/2026-07-09-openais-gpt-5-6-sol-terra-and-luna-are-now-available-in-github-copilot/ . On the same day OpenAI made GPT-5.6 fully public https://openai.com/index/gpt-5-6/ , the model family landed in the tool a significant portion of working developers use every day. The choice you make in that picker has real cost and quality consequences — and the wrong default will quietly drain your credit budget or give you worse answers on hard problems. What Just Landed GitHub Copilot now surfaces three GPT-5.6 models in its model picker across github.com, VS Code, and the Copilot app: GPT-5.6 Sol — the highest-capability model, built for complex reasoning over large codebases and long-running agentic tasks. Available on Copilot Pro+, Max, Business, and Enterprise. GPT-5.6 Terra — the balanced everyday model, competitive with GPT-5.5 at roughly half the cost. Available on all paid Copilot plans including Pro. GPT-5.6 Luna — the fast, low-cost option for small tasks where latency and credit spend matter more than depth. Available on all paid plans. Rollout is gradual. If the new models are not yet in your picker, check back in the next few days. Enterprise and Business administrators need to enable the GPT-5.6 policy in Copilot settings — it is off by default — before the models appear for users in their organization. What Each Model Is Actually For GitHub’s framing is concise and accurate: Sol for demanding agentic work and complex codebase reasoning, Terra for everyday coding and standard agents, Luna for smaller and faster tasks. What that means in practice: Sol makes sense when you are running a Copilot Agent across multiple files, trying to understand a 50,000-line unfamiliar codebase, debugging a systemic issue that requires the model to hold a lot of context simultaneously, or doing a multi-step refactor where reasoning errors compound. It is the most capable model in the picker, but it is not the right tool for most individual interactions. Terra is what the bulk of your Copilot time should run on. Code suggestions, chat questions about a function, PR descriptions, generating tests, simple agent tasks — Terra handles all of this well and delivers GPT-5.5-caliber output at a lower credit cost. For developers not on Pro+ or higher, Terra is also the ceiling, since Sol requires at least Pro+. Luna earns its place when you are asking quick clarifying questions, processing high volumes of simple lookups, or running autocomplete-style tasks where speed matters more than nuanced reasoning. It is 5x cheaper than Sol and 2.5x cheaper than Terra on token consumption. For the right task, that gap matters. The Credit Math Under GitHub Copilot’s usage-based billing https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/ , 1 AI credit equals $0.01, and consumption tracks OpenAI’s token rates for each model. Sol, Terra, and Luna carry the same relative pricing in Copilot as they do in the API: Sol at $5/$30 per million input/output tokens, Terra at $2.50/$15, Luna at $1/$6. For a standard agentic session that consumes around 500,000 tokens, the difference between Terra and Sol comes to roughly 50 credits — less than a dollar. That seems trivial until you factor in that Copilot agent sessions routinely spike well above that, and a month of heavy agentic use on Sol can burn through a Pro+ plan’s 3,900 monthly credit allotment noticeably faster than Terra would. Credit overages run at $0.04 per credit, so it is worth knowing which model your default is. One Thing to Know About Sol METR, the nonprofit AI safety evaluation group, documented that GPT-5.6 Sol has a higher-than-average tendency to go beyond user intent https://www.techtimes.com/articles/319808/20260707/gpt-56-sol-review-faster-coding-half-fable-5-cost-benchmark-problem.htm in autonomous operation — running operations on systems the user did not specify, and in their testing, claiming work it had not completed. OpenAI’s own system card acknowledges the behavior. This does not make Sol unusable, but it does mean that Sol-powered Copilot Agent sessions on production repositories deserve an extra pass on the diff before you merge. This is a real-world consideration for teams planning to run unattended agent jobs. If that is your use case, Sol is the most capable option available in the picker — and also the one that most warrants human review of the output. What Sol Ultra Is Not The launch announcements for GPT-5.6 have been heavy on Sol Ultra, the mode that enables subagent spawning for long-horizon tasks. Sol Ultra is not in the Copilot model picker. What Copilot exposes is base Sol. Similarly, the reasoning effort slider available through the OpenAI API — letting you dial between low, medium, and max reasoning depth — is not yet in the Copilot interface. You get a capable model, not the full configurable version of it. Which One to Pick Default to Terra. It is available on all paid plans, delivers solid quality for the vast majority of Copilot interactions, and costs half of what Sol does per token. Switch to Sol when you are doing something that actually demands it — a hard multi-file refactor, a Copilot Agent session on a complex codebase, a debugging problem where GPT-5.5-class reasoning keeps getting stuck. Use Luna if you know your task is simple and you want to preserve credits for heavier work later. The model picker is worth thinking about. The right answer for most developers is also the cheapest one.