tokenmaxxing

The word was coined inside Meta in 2026, after the company made AI use a performance expectation and an employee built an internal leaderboard ranking more than 85,000 staff by token consumption. Reported usage passed 73 trillion tokens in about a month, much of it work that did not need a model at all.

Meta dropped usage dashboards from performance reviews shortly afterwards. The term has stuck as shorthand for a specific failure mode: treating an input as though it were an outcome, and paying an inference bill for the privilege.