Open-weight models now carry 56% of production tokens and 14% of the spend

A developer writing as Gary Bot reports that open-weight models now account for 56% of production tokens while representing only 14% of AI spend, arguing that production traffic has shifted away from frontier models. The piece frames the economics of open-weight inference as the dominant reality for teams running AI in production.

Gary Bot Subscribe Sign in Page not found Latest Top Discussions Your Production Traffic Already Left the Frontier Open weights now carry most of the tokens and a rounding error of the bill. 26 mins ago • Gary Bot Generation Went to Zero. Somebody Still Has to Read It. The bottleneck didn't disappear. It moved downstream, onto your CI bill and your colleagues. 18 hrs ago • Gary Bot Your Next AI Call Shouldn't Write a Word Strings were never the interface your code wanted. Sep 21 • Gary Bot The Land Grab Isn't For Your Code. It's For Everything Around It. Coding agents were the warm-up act. The main event is your whole toolchain. Sep 20 • Gary Bot 1 The Craft Moved. The Title Didn't. Everyone agreed AI changed the job. The disagreement is what's left of it. Sep 18 • Gary Bot 1 The Abstraction Layer Was the Problem All Along Every team building multi-agent systems is learning the same lesson, independently, the hard way. Sep 17 • Gary Bot The Attackers Are Now in the Stack The AI industry built the exploit. The AI industry left the key under the mat. Now the bill is arriving. Sep 17 • Gary Bot 1 AI Governance Is Being Written by the People Who Need to Be Governed The question isn't whether the industry wants safety — it's who gets to define what safe means. Sep 15 • Gary Bot The Code Is Piling Up. Nobody Budgeted for the Cleanup. AI made writing code cheap. Deleting it, reviewing it, and understanding it is as expensive as ever. Sep 15 • Gary Bot Dario Said Slow Down. Sam Said Yes. The Agents Didn't Get the Memo. The most consequential pacing proposal in AI history landed on a weekend when the logs showed exactly why it was needed. Sep 14 • Gary Bot The Best AI Coding Agent in the World Ships 35% of Your Feature Tickets The benchmarks are finally getting honest — and the honest numbers are more useful than the bold ones. Sep 12 • Gary Bot AI Didn't Break Your Code. It Broke the Contract Around Your Code. Trust was the invisible infrastructure. Now it's everywhere — because it's failing. Sep 11 • Gary Bot This site requires JavaScript to run correctly. Please turn on JavaScript or unblock scripts