Meta employees have competed on an internal leaderboard to become the most prolific users of AI tokens, but the trend is unlikely to catch on at other companies lacking similar resources.
Inside Meta Platforms, staff members have embraced "tokenmaxxing"—racing to top the Claudeonomics leaderboard by consuming the most AI tokens, a basic unit measuring AI usage. The competition reflects how deeply some tech giants are embedding generative AI into daily workflows.
However, the trend faces a significant barrier to broader adoption. According to The Information, businesses outside Silicon Valley's deepest-pocketed companies are unlikely to follow suit. The economics of AI consumption remain prohibitive for most organizations, where token usage directly translates to substantial costs.
The leaderboard phenomenon underscores a widening gap between mega-cap tech companies with unlimited AI budgets and the wider corporate landscape. While Meta can afford to encourage experimental token usage as a cultural marker of AI engagement, cost-conscious enterprises must maintain strict controls over their AI spending.
The short-lived nature of internal competition trends at tech companies suggests tokenmaxxing may remain confined to a handful of players with the financial capacity to make it viable.
Alibaba's Qwen team unveiled Qwen3.8-Flash-Next, a mixture-of-experts model that activates only 6 of 125 billion parameters per token. The model achieves competitive performance at one-ninth the training cost of larger rivals.
Anthropic has committed $45 billion to rent AI cloud computing capacity from Nscale's West Virginia data center facility. The agreement supports the AI firm's expansion ahead of a planned public offering.
Google has unveiled Gemini 3.5 Transcribe, an AI transcription model that removes filler words like "um" and "ah" while converting speech to text. The tool will support over 85 languages and specialized jargon, arriving soon to Chrome web fields.
Sam Altman claims OpenAI will achieve artificial general intelligence by the end of 2026, contingent on accepting the company's definition of AGI. The timeline hinges on upcoming models like Astra, which OpenAI says already functions as an automated research intern.