Meta’s ‘Claudeonomics’ Data
Inside Meta, an internal 'Claudeonomics' leaderboard showed top users averaging roughly 281 billion tokens per month and the company reporting about 60 trillion tokens used in 30 days, illustrating how teams can gamify and massively scale model consumption. Those numbers signal that token usage can quickly become a meaningful operational and cost metric to present in engineering reviews. (x.com)
Meta employees were reportedly competing to become “Token Legend” on an internal leaderboard that ranked people by how many artificial intelligence tokens they burned, and the dashboard showed more than 60 trillion tokens used across the company in 30 days before Meta took it down. (theinformation.com) (gizmodo.com) A token is the billing unit for a language model, the way kilowatt-hours are the billing unit for electricity, and companies pay for both the words they send in and the words the model sends back. OpenAI’s developer docs describe tokens as the chunks models process, and Anthropic’s pricing page lists prices per million tokens for Claude models. (developers.openai.com) (platform.claude.com) That is why the Meta numbers startled people. One reported screenshot showed the top user averaging about 281 billion tokens in a month, which works out to roughly 9.4 billion tokens a day. (theinformation.com) (finance.biggo.com) The leaderboard was called “Claudeonomics,” a wink at Anthropic’s Claude model, and it reportedly tracked usage across more than 85,000 users while listing the top 250 inside Meta. That turned model consumption into something visible, comparable, and easy to brag about. (theinformation.com) (kucoin.com) Once a company posts a ranking, people start optimizing for the ranking. Reports said some employees were leaving agents running just to push their totals higher, the same way a step counter can turn a walk into a contest. (blog.tmcnet.com) (the-decoder.com) The cost side is why executives will pay attention. Anthropic’s current price sheet shows Claude Opus 4.6 at $5 per million input tokens and $25 per million output tokens, so even rough back-of-the-envelope math turns trillions of tokens into a line item large enough to show up in engineering reviews. (platform.claude.com) Using the cheapest listed Claude Opus 4.6 input rate alone, 281 billion tokens would imply more than $1.4 million for one month of input tokens for one heavy user, and real bills can move higher once output tokens and cache charges are included. That estimate is an inference from Anthropic’s posted pricing, not a statement of what Meta actually paid. (platform.claude.com) (dnyuz.com) Meta’s scale makes that easier to imagine. The company’s latest annual report filed in January 2026 shows Meta is operating at enormous size, so a tool that nudges tens of thousands of employees toward heavier model use can move spending fast. (sec.gov) The twist is that the leaderboard did not last. Multiple reports on April 9, 2026 said Meta shut Claudeonomics down after internal usage data began circulating publicly. (theinformation.com) (indiatoday.in) What survives the shutdown is the metric. In software teams now, “How many people used the model?” is turning into “How many tokens did this workflow consume, and what did we get for them?” (theinformation.com) (platform.claude.com)
