High AI Users Consume 10x More Tokens but Deliver 77% Higher Pull Request Throughput

The top 10% of Claude Code users consumed about 10 times as many AI tokens as the median developer but produced only about twice the output, according to data compiled by Jellyfish. Weekly Claude Code consumption for top AI adopters reached 225 million tokens per user. The median software engineer consumed 32 million tokens per week.

AI tokens are small chunks of text that AI models break words and inputs into to process them and serve as a way to price AI usage, usually at a cost per million tokens. Very high-adoption teams posted 77% more pull request throughput than low-adoption teams. Jellyfish has access to hundreds of companies and the coding behavior of hundreds of thousands of software engineers.

Nicholas Arcolano, Jellyfish's head of AI and research, said that the gap between token consumption and output is the clearest sign that extreme "tokenmaxxing" is not a sustainable strategy. "CFOs are getting on people's case," he said. " Arcolano added that raw token volume is too messy to use on its own as a productivity score.

AI model changes can cause token counts to swing dramatically even when behavior does not change, meaning a developer's token spend is not always a reliable proxy for true productivity. Instead, leaders should track cost per pull request or another outcome-based metric, not token totals, Arcolano said.

"Even if you rationalize that you're getting more value from these things than you would get from a human doing the same work, if token costs surge, a CFO will still worry that you broke their spreadsheets," he told Insider.

The findings come as the tech industry enters a new phase of AI spending discipline. Heavy AI token consumption yields limited returns while increasing costs for companies. Arcolano described one pattern among extreme users: "Rather than think ahead about the right way to do something, I'll have five AI agents build it five different ways and pick the winner.