Meta Kills Claudeonomics Leaderboard; Final Champions Were Asking Claude How To Use More Claude

MENLO PARK—Hours before Meta CTO Andrew Bosworth retired the company's internal "Claudeonomics" token leaderboard Tuesday, records reviewed by The Artifice indicate that a significant portion of the 6,000 participating engineers spent their remaining unrestricted access asking Claude, Anthropic's competing AI model, how best to generate more tokens before the deadline.

"Write 500 distinct, thoughtful prompts I can submit to an AI assistant over the next four hours," one entry, timestamped 8:14 p.m., reportedly read. The resulting output — 1.2 million tokens of professionally formatted questions spanning philosophy, logistics, and competitive coding — was fed back into Claude in its entirety, generating an additional 3.4 million tokens of responses, which were then submitted as new prompts.

Bosworth's memo, which warned that Meta's AI spend was "heading into the billions" after employees consumed 73.7 trillion tokens in a single 30-day window, did not specify whether the terminal sprint had been included in the total.

The leaderboard, named after Anthropic's flagship model despite Meta's active development of its own Llama series, had awarded no prize beyond top-10 placement on a dashboard the memo described as having "created some really exciting energy." Beginning in 2027, that energy will be redirected toward MetaCode, Meta's proprietary AI system, which internal documentation describes as "available at no additional external cost."

The company declined to say how many tokens were spent determining the optimal way to spend tokens.

"There's something poetic about it," said a former member of Meta's AI safety team, who departed in February. "The last thing they did with unlimited Claude access was ask Claude what to do with unlimited Claude access. I don't know what else you'd expect."