Gemini 3.7 Flash became selectable in Copilot on 13 August. The plan list, the surface list and the admin requirement are all now entirely predictable — which makes the one unusual detail worth pulling out.
The policy administrators have to enable is called the Gemini 3.7 Flash Preview policy. Not the Gemini 3.7 Flash policy. Preview is doing work in that sentence.
Why "Preview" in the policy name matters
A preview model in a production coding tool is a different proposition from a generally available one. Preview status typically means the model, its behaviour or its availability can change without the notice a GA product carries, and it means the thing your team standardised on last week may not be the thing they get next month.
That is fine for evaluation and risky as a default. The distinction worth drawing inside your own team is between "you may select this" and "this is what we use" — a preview model belongs firmly in the first category.
It also has a practical consequence for anyone writing internal documentation: do not put a preview model name in a runbook. Put the decision procedure in the runbook and let the model name live somewhere it is cheap to change.
What GitHub says it is good at
Early testing is cited across six areas, and the list is broader than the usual "good at coding":
- Web and app development
- Agentic coding workflows
- Code quality
- Output presentation — how readable and well-organised the response is
- Codebase research
- Verification during complex coding tasks
Where and how you get it
Available on Pro, Pro+, Max, Business and Enterprise, in VS Code, Visual Studio, the Copilot CLI, the Copilot cloud agent, the Copilot app, JetBrains, Xcode and Eclipse.
Business and Enterprise administrators must enable the Gemini 3.7 Flash Preview policy in Copilot settings before anyone in the organisation can select it. Billing is provider list pricing under usage-based billing; GitHub states no premium request multiplier.
The rollout is gradual, as with every model this month.
A Flash model is a routing decision, not a replacement
Flash-class models exist to be fast and cheap, and the mistake teams make is treating the arrival of one as an argument about which model is best. It is not — it is an argument about which work goes where.
The useful pattern is the same one that applies to MAI-Code-1.1-Flash, which reached Copilot two days earlier: send the routine volume to the small fast model, keep the expensive model for the work that actually defeats it, and make sure someone notices when the cheap one is failing rather than merely finishing.
That last clause is the whole discipline. A fast model that produces plausible wrong answers quickly is more expensive than a slow model that produces right ones, and no billing report will tell you which you have.