Microsoft’s MAI-Code-1.1-Flash is now in production in GitHub Copilot. GitHub’s model pricing documentation lists the Microsoft-built coding model as generally available, with rates of $0.20 per million input tokens and $1.20 per million output tokens.
In its October 7 announcement, Microsoft says the model produces better code at one-quarter the price of the version it launched in June. It also reports faster token streaming, lower token consumption per task, and improvements in command-line and.NET work.
Those claims make this LLM release worth testing in an existing development workflow: do the lower rates and claimed efficiency translate into cheaper, faster changes that pass review? Microsoft’s figures give developers a reason to try it, without establishing that it should replace their current model across every task.
Production Copilot Access Is Separate From Local Access
Hosted production access is available now: Microsoft says MAI-Code-1.1-Flash is already running in GitHub Copilot, and GitHub lists it as generally available in its pricing table.
The announcement describes a separate on-device release. Microsoft says the model is available to download and run locally, while experimental local-model access through the GitHub Copilot app, GitHub Copilot CLI, and Visual Studio Code is scheduled for the end of October. That timeline applies to experimental on-device integration; hosted production availability was announced on October 7.
Running the local version comes with a substantial hardware recommendation. Microsoft recommends devices with more than 120GB of RAM for best performance. It says 3-bit quantization reduces memory requirements while preserving a 256K context window and comparable performance to the full-precision version on the named coding benchmarks.
Microsoft describes local model calls as carrying zero inference charges. This covers local inference billing, with no promise of cost-free hardware or equivalent performance on an ordinary development laptop. For developers evaluating the shipped Copilot change, hosted usage is the more immediate comparison.
Published Token Rates Make the Cost Claim Testable
Microsoft’s quarter-price claim uses its June model as the comparison. It does not promise that a Copilot subscription, or an organization’s entire Copilot bill, will fall by 75%.
GitHub’s documentation provides the usage rates:
