Chinese open-source models dominate OpenRouter charts; DHH switches to Kimi K2 as daily driver
Feb 18, 2026
Key Points
- Chinese open-source models dominate OpenRouter's rankings, with 37signals founder David Heinemeier Hansson adopting Kimi K2 as his daily driver for production work.
- OpenRouter captures only 1–2% of global LLM token volume and skews toward open-source by design, making its rankings unrepresentative of broader market adoption.
- Open-source model commoditization is real and competitive enough to displace closed alternatives at lower cost, independent of whether these models matter to frontier AI competition.
Summary
Chinese open-source models top OpenRouter's public rankings. David Heinemeier Hansson, founder of 37signals, switched to Kimi K2 as his daily driver for bug fixes, showing real adoption beyond the API aggregator's charts.
OpenRouter captures only 1–2% of global LLM token volume. The aggregator skews toward open-source models by design because developers using proprietary APIs hit those providers directly rather than routing through a middleman. Major labs do not publish inference through OpenRouter, so the platform's rankings reflect a narrow slice of usage, not the broader market.
Consumer LLM products generate far more tokens. Google's AI Overviews embedded in search results likely represents the largest LLM inference in the world. Gemini, Claude, and other consumer apps bypass OpenRouter entirely. Chinese open-source models commanding OpenRouter's top positions and a respected developer like Hansson trusting one for production work suggests these models are competitive enough for real tasks at lower cost.
Open-source model commoditization is real, even if OpenRouter's visibility into it is incomplete. Whether open-source models matter to frontier AI competition is separate from whether they work well enough to displace closed alternatives at scale.