Microsoft asks its engineering teams to rethink how they use generative AI.
In an internal memo, disclosed by 404 Media, Jay Parikh, Executive Vice President of CoreAI, states that the “tokenmaxxing” is not the goal and urges teams to prioritize the value created per token.
Thus, Microsoft has set OpenAI’s GPT-5.6 Sol as the default model in GitHub Copilot used by its employees. This setting is not mandatory, however. Microsoft engineers may choose another model when they believe it is more suitable for a specific task.
This shift concerns only internal Microsoft usage and does not, at this stage, change the model choices offered to GitHub Copilot customers.
GPT-5.6 Sol becomes the default model for Microsoft collaborators
According to internal documents reviewed by 404 Media, each Microsoft division has an objective for AI token consumption. Engineers can track their own usage via an internal dashboard. The monthly token expenditure per engineer typically ranges from a few hundred to several thousand dollars, depending on teams and projects. Microsoft has not published official figures regarding these costs.
According to Jay Parikh, the aim is not to slow down Microsoft’s AI-first strategy but to treat tokens like any other computing resource, taking their cost into account and seeking the best possible return on investment.
This shift comes as the costs associated with inference become a major concern for AI service providers. As companies broaden the use of development assistants, copilots, and AI agents, the ability to maximize value produced per token becomes a critical lever for cost control.
The fact that GitHub Copilot now falls under the CoreAI engineering group, led by Jay Parikh alongside Visual Studio and Visual Studio Code, explains why these new guidelines apply primarily to that development environment.
