Meta's internal "Claudeonomics" leaderboard revealed, 85k employees burned through 60 trillion tokens in 30 days

ME News Report, April 7 (UTC+8), an AI usage leaderboard named “Claudeonomics” has appeared on Meta’s internal network, named after Anthropic’s flagship product Claude. It was built by employees using company data, aggregating token consumption from over 85k people and listing the top 250. A copy of the leaderboard seen by The Information shows that in the past 30 days, total consumption exceeded 60 trillion tokens. Based on Claude Opus 4.6’s publicly available average price (about $15 per million tokens), this roughly amounts to $900 million, though Meta’s actual model combinations and protocol prices are unknown. The top individual user consumed an average of 281 billion tokens, with costs possibly reaching several million dollars.

The leaderboard features gamified incentives, ascending from bronze to emerald levels, with top titles including “Token Legend” and “Session Immortal,” as well as “Model Connoisseur” and “Cache Wizard.” Some employees push their AI agents to run research tasks continuously for hours solely to boost their usage numbers. Meta CEO Mark Zuckerberg and CTO Andrew Bosworth did not make it into the top 250.

A recent trend in Silicon Valley called “tokenmaxxing” has emerged, with token consumption becoming a new metric for measuring engineers’ productivity. Bosworth stated at a tech conference in February that a top engineer’s token expenditure had become comparable to their salary, boosting productivity by up to ten times. “This is a guaranteed profit, keep burning tokens, no upper limit.” Nvidia CEO Jensen Huang last month said that if an engineer earning $500k annually spends less than $250k on tokens each year, he would be “deeply alarmed.”

Meta engineers currently use external models from Anthropic, OpenAI, and Google, as well as internal tools like MyClaw (Meta’s version of OpenClaw) and the recently acquired Manus. In an internal memo this year, Zuckerberg issued a “bold request” to engineering teams: rewrite Meta’s codebase so that AI agents can directly read and modify code. (Source: Meta))