OpenAI has revealed the staggering costs of its researchers’ reliance on AI coding agents. In a blog post, the company said median researchers now spend over $600 a day in tokens, while top users burn through more than $7,000 daily.
Tokens — the small text chunks an AI model reads and produces — have become central to how OpenAI staff build, test, and ship code.
Coding Agents Transform Software Engineering at OpenAI
CEO Sam Altman previously noted that the company’s highest token user consumes 100 billion tokens a month. OpenAI said usage has surged in recent months, with coding agents increasingly handling high‑level tasks and troubleshooting, reducing human technical support requests by half since January.
The company claims it has achieved Altman’s goal of building an “automated research intern” and is on track to create a fully automated AI researcher by 2028.
The rise of coding agents has transformed software engineering, with many developers delegating tasks to AI. Yet the trend has also sparked concerns over “tokenmaxxing,” where employees use AI excessively without productivity gains.
Companies like Meta and Amazon have shut down internal leaderboards that encouraged heavy usage, warning staff against using AI “just for the sake of it.” OpenAI, however, continues to embrace high token consumption as part of its research culture.
