The News, Explained

GitHub added Microsoft’s MAI Code 2 and MAI Code 2 Flash coding models to Copilot on November 17. Code 2 is available on Pro, Pro+, Max, Business and Enterprise; Flash also reaches Free and Student. On Free and Student, Flash can be used through Auto model selection, where Copilot picks the model. Eligible annual subscriptions qualify, and Business or Enterprise admins can control access through their model policies. Source Source

GitHub positions Code 2 for work spanning several files or requiring a view of the wider codebase. It presents Flash as a lighter option for explanations and focused edits. Explaining how one function works, for instance, calls for a different scope of review from changing a shared function’s arguments and updating its callers and tests. The scope of the change gives developers a practical way to decide which model to try first. Source

When extra usage beyond a plan’s allowance is billed, GitHub calculates AI credits from the token rates. One credit is worth US$0.01. For a million regular input, cached input and output tokens, respectively, the launch rates are US$0.50, $0.025 and $2.50 for Code 2, and $0.05, $0.005 and $0.25 for Flash. All six rates double starting May 17, 2027. Inline code completions and next edit suggestions in the editor are excluded from AI credit billing. Source

OYOPICK’s Take

For a short explanation or a small edit, it makes sense to check what the lighter model can do before moving up to work that spans several files. During the promotion, Flash’s output rate is one tenth of Code 2’s for the same number of output tokens. Repeated attempts can change the cost of the job. Teams could track the tokens used and the corrections people still made on tasks of similar size. That would make it easier to choose a model and plan for the end of the discount. Source Source