On Monday, Decagon CEO Jesse Zhang posted a provocative new theory posted under the headline. “Everybody Gets It Wrong About Open Source AI in the Enterprise.” The post tackles one of the most interesting contradictions in today’s AI economy: More mature AI deployments are shifting to lighter models, he says, even at his own company. But the total cost of expensive modern models has almost decreased.
This is a new way of thinking about the relationship between frontier and open source models. According to Zhang, they are not competitors, and the success of open source models does not come at the expense of frontier labs. Instead, they are two stages of the same lifecycle, expensive frontier models used to prove use cases that can be transferred to cheaper open source alternatives as they mature.
Like more mature use cases switch to lighter modelsnew use cases continue to emerge – and the overall cost of frontier models is almost always reduced.
Zhang doesn’t provide much information to support the point, but the data isn’t hard to find. Vercel’s AI gateway dashboard shows that over the past week, DeepSeek has risen to the top in terms of token volume, now processing just over a third of the tokens passing through the company’s infrastructure. Z.ai – the lab behind the popular GLM-5.2 model – rose to a respectable fourth place during the same period.
But if you move on to total token spending, you’ll see that Anthropic still accounts for more than half of the total AI spending on the platform. Given that much of the recent change has come from Anthropic’s own rising prices, the stock has dipped slightly over the past month, but not significantly.
OpenRouter tells a similar story, capturing a larger (but slightly less enterprise-y) segment of the market. DeepSeek V4 Flash is the main winner in total usage, processing 5.3 trillion tokens weekly. Opus 4.8, the most popular frontier model, manages just over 2 trillion. OpenRouter doesn’t rank the models by total spend, but it lists the average token value for Opus 4.8 as roughly 23 times higher than V4 Flash ($1.37 per million tokens compared to just 6 cents), meaning Opus is still likely capturing the share of spend.
Those numbers don’t even include Nvidia’s newest arrival, the Nemotron. preparing to jump to the front of the pack Thanks to Nvidia’s strong connections and the extreme adaptability of the model itself.
These numbers don’t quite prove Zhang’s point about AI lifecycles, but they do show that frontier labs like Anthropic aren’t suffering much from the rise of open source—at least not yet. One explanation is that the market for AI-addressable tasks is growing so fast that top models can only maintain their positions by favoring early-stage deployments. As Zhang says, “Frontier labs will continue to own discovery. Open source will own more and more production.” Another explanation could be that even if customers switch to open source, many use cases are so difficult that they cannot be completely replaced by cheaper alternatives.
Either way, this two-tier model economy may become a relatively stable feature of the AI economy.
Last September, I wrote about the possibility of the end of the foundation labs He sells coffee beans to Starbucks — that is, the application layer serves as a commodity input while benefiting. Parts of this prediction came true: Vertical AI games, for one, moved to lighter models, and the economics of “GPT wrapper” startups remained largely stable.
But we also see that token-to-token, frontier providers have been able to retain the most desirable part of the market – the premium token price. And that doesn’t seem likely to change anytime soon.
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
