Meta Embarrassed: Its Employees Prefer Anthropic's AI

Meta finds itself in a delicate situation after an internal ranking revealed that its employees prefer to use competing artificial intelligence models. This ranking, named "Claudeonomics," was established by a Meta employee and was disclosed by The Information. It was named after Claude, the model from Anthropic, a direct rival of Meta in the AI space.

This table listed the 250 largest AI consumers within the company over a 30-day period, involving more than 85,000 employees. The leader of the ranking alone consumed 281 billion tokens, and in total, employees burned through 60 trillion tokens. Badges such as "Token Legend," "Cache Wizard," and "Session Immortal" rewarded the most active users.

External Models at the Top

The choice to name the ranking "Claudeonomics" is telling: Meta engineers prefer using Claude from Anthropic, GPT from OpenAI, and Gemini from Google, rather than internal models like Llama. This phenomenon can be attributed to the delay of Llama 4, a reorganization of the AI teams, and the departure of many researchers, leaving a void filled by competing tools.

Following the leak of this information to the press, Meta quickly removed the ranking, explaining that it was "a fun way to look at tokens." Management made this decision after the data was shared externally.

The "Tokenmaxxing" Fever

This phenomenon is not limited to Meta. Jensen Huang, CEO of NVIDIA, stated that a well-paid engineer should consume a significant amount of tokens. Shopify, for its part, is also tracking its employees' consumption, valuing those who use the most tokens. However, this practice raises questions, as it measures data input rather than actual productivity.

At Meta, some employees even let AI agents run unnecessarily to improve their ranking position. The theoretical cost of these 60 trillion tokens, at Claude's public rate, could reach $900 million, although the actual price is likely lower due to negotiations.

Intensive use of AI also leads to side effects, such as "AI Brain Fry," a cognitive fatigue and loss of concentration among the most diligent users.

Meta recently unveiled Muse Spark, the first model from its new elite AI team, but it remains to be seen whether its own employees will adopt it.