Microsoft sets AI token budget targets, adopts GPT-5.6

Microsoft has implemented new company-wide restrictions on employee use of artificial intelligence tools, introducing departmental budget targets and a cheaper default AI model in an effort to control costs while emphasizing business outcomes over raw consumption. The policy, communicated to staff in an internal email from Executive Vice President Jay Parikh, aims to curb a practice informally known as "tokenmaxxing," where employees maximize AI usage without regard for expense.

Under the new guidelines, which took effect in July 2026, all Microsoft divisions now operate with an "AI token budget target." Employees can track their individual AI spending, though specific monetary limits have not been publicly disclosed. Internal data indicates many engineers currently spend hundreds to several thousand dollars per month on AI tokens, which are units of consumption that increase with the volume of text processed by AI models.

A central component of the strategy is the adoption of OpenAI's GPT-5.6 as the default AI model for internal use. Company leadership stated this model was selected because it is cheaper than several alternatives, allowing Microsoft to get greater value from...