MAI Code 2 vs Flash on the same workload

Compare published Copilot rates, separate input from output, and measure the cost of a completed coding task.

Price the same workload first

During the promotional period GitHub lists MAI Code 2 at $0.50 for a million input tokens, $0.025 for a million cached input tokens and $2.50 for a million output tokens. The corresponding Flash rates are $0.05, $0.005 and $0.25. GitHub schedules a doubling of each of those rates on May 17, 2027. GitHub pricing tables and effective dates

For 1,000 calls with 4,000 uncached input, 8,000 cached input and 1,500 output tokens each, the promoted MAI Code 2 cost is $5.95. The same input and output volumes cost $0.615 on Flash. The fivefold difference on cached input and tenfold differences on the other categories make the overall ratio depend on your usage mix.

Compare the cost of a completed task

Choose representative issues with known acceptance criteria, such as passing a set of existing unit tests and preserving a documented response format. Run the same starting state and specification on each model. Record total model usage across attempts, the number of accepted changes, and any manual rework. Divide the total usage cost by the number of accepted changes only when at least one task passed. A cheaper individual call can still require more calls to finish a task.

Microsoft describes Flash as the smaller option for everyday coding and publishes evaluation conditions with the launch. Those results concern its selected benchmarks. The per-task cost in your own repository requires your own token counts and pass/fail results. Microsoft model announcement

Estimate a Flash workload or prepare your input figures.