Open Source AI Models Are Rising, But Frontier Labs Like Anthropic Aren't Hurt Yet
Despite the rapid growth of open source AI models, expensive frontier models still capture the majority of spending, according to data from platforms Vercel and OpenRouter.
Decagon CEO Jesse Zhang has published a theory explaining a seemingly contradictory situation: as mature AI deployments shift to lighter models, spending on frontier models remains steady. He argues that open source and frontier models are not competitors but two phases of the same lifecycle – expensive models are used for early-stage discovery, while cheaper alternatives take over in production. As mature use cases move to lighter models, new use cases emerge, keeping frontier spending nearly unchanged.
Vercel's data shows that in the past week alone, DeepSeek has surged to the lead in token volume, processing over a third of tokens on the platform. However, when looking at overall spending, Anthropic still accounts for more than half of AI expenditures on Vercel. While this share has dropped slightly in the past month due to Anthropic's own price increases, the decline is not significant.
OpenRouter, covering a broader market segment, tells a similar story. DeepSeek V4 Flash leads in usage, processing 5.3 trillion tokens weekly, while the most popular frontier model, Opus 4.8, handles just over 2 trillion. But Opus 4.8 costs roughly 23 times more per token ($1.37 vs. $0.06 per million tokens), meaning it likely still captures the lion's share of spending.
Zhang suggests that frontier labs will continue to own discovery, while open source will increasingly own production. Another explanation is that some use cases are so difficult that they cannot be fully replaced by cheaper models. Although some predicted that frontier models would become commoditized, the data shows they have maintained premium token prices. Nvidia's upcoming Nemotron model, leveraging Nvidia's strong connections and adaptability, is poised to become a new leader.
