Facebook Shuts Tokenmaxxing Scoreboard After Developers Burned 30% More AI Tokens
Updated
Updated · InfoWorld · May 12
Facebook Shuts Tokenmaxxing Scoreboard After Developers Burned 30% More AI Tokens
1 articles · Updated · InfoWorld · May 12
Facebook pulled an internal dashboard that ranked developers by Claude Code token consumption, ending a contest that crowned top users with labels such as “Token Legend” and “Cache Wizard.”
The company moved after the metric encouraged “tokenmaxxing” — developers reportedly used tools like OpenClaw to chew through large volumes of tokens without producing better software.
Token burn was easy to count and display, but the report argues it worked like old lines-of-code targets: a gamable measure that can reward waste and lower-quality output.
Facebook acted quickly partly because such dashboards can attract executive attention and harden into OKRs or investor-facing metrics, even though higher token throughput says little about engineering performance.
Is 'tokenmaxxing' a billion-dollar blunder, or a secret strategy for training next-gen AI?
If AI token counts are a flawed metric, how can companies prove their trillion-dollar AI investments are actually working?