Particle.news
Download on the App Store

Microsoft Tells Engineers to Stop ‘Tokenmaxxing’ and Sets GPT‑5.6 Sol as Copilot Default

The company says the move is meant to raise value per token by routing work to OpenAI’s model and giving divisions AI token budget targets.

Overview

  • Microsoft made OpenAI’s GPT‑5.6 Sol the default model for internal GitHub Copilot use on Wednesday and told staff to prefer outcomes over raw token use.
  • Jay Parikh’s internal memo instructs engineers to stop “tokenmaxxing,” defines tokens as the units that drive AI cost, and links to updated Copilot guidelines for tracking spending.
  • Since July, each Microsoft division has carried an AI token budget target and employees can view their personal token spend, though CoreAI has not imposed team or individual caps.
  • The default change shifts large volumes of internal coding traffic away from Anthropic’s Claude models and is framed as capturing more value from Microsoft’s commercial ties with OpenAI.
  • The decision reflects a wider industry shift to AI FinOps where firms use model routing, budgets, and spend visibility to curb runaway inference bills while keeping AI-driven product goals moving forward.