The AI Market: Why the "Token" Metric is Misleading and Why You Should Count in Dollars
Recently, it has become fashionable in the artificial intelligence industry to compare models by the volume of tokens consumed. However, as recent discussions among leading analysts show, this approach is fundamentally flawed. Dragonfly venture firm managing partner Haseeb Qureshi rightly points out that token share is an extremely unreliable metric for assessing the AI model market. The real picture of market strength and demand is provided only by dollars spent, not raw consumption.
Why does the token metric distort reality so much? Qureshi highlights several key problems. The first is subsidies. Chinese labs regularly launch new models with aggressive discounts or even free access. This artificially inflates token consumption volumes, as users migrate from one free model to another without spending a cent. The charts show rapid growth, but there is no real monetization behind it.
Size Matters: Price per Token
The second problem lies in the different sizes of models. Small models, such as Qwen 3.5-27B, can be a hundred times cheaper per token than flagship solutions like Claude Opus. A sharp spike in Qwen's consumption share on OpenRouter charts looks like a triumph for open models, but in economic terms, it is negligible. Correct analysis, in my deep conviction, is only possible within weight categories—models should be compared within the same class by size.
The third problem is multi-agent systems. You can spend the same amount on a complex configuration based on DeepSeek or GLM 5.2 and on a single request to the advanced Opus model. However, a multi-agent architecture will "burn" several times more tokens for the same money. Qureshi provides a clear example: if 5% of Opus usage migrates to such a system with a fourfold token consumption, the chart will show an 18% loss in Opus's share, even though actual spending shifted by only 5%. This is a classic example of how the token metric exaggerates the importance of cheap tokens.
OpenRouter: A Distorted Mirror of the Market
The fourth problem is related to the choice of the OpenRouter platform itself. Large companies, having settled on one advanced lab, prefer to work directly with Anthropic or OpenAI, bypassing OpenRouter's markup. On the chart, this looks like a decline in the US share, even though the tokens are simply moving off the platform. The conclusion is obvious: OpenRouter is useful for assessing the share within open models but is absolutely unsuitable for comparing open and closed systems.
This position is further developed by SageRoad Research founder Trevor Noren, linking it to pricing pressure on the industry. JPMorgan data shows that many tokens in the future will be consumed not by advanced models, but by small open models sufficient for specific tasks. Amazon already offers about fifty open models at a price that is a fraction of the cost of advanced ones.
JPMorgan's figures are particularly telling: running the Artificial Analysis Intelligence Index benchmark on Claude Opus 4.8 costs $3,700 for a score of 56, while DeepSeek V4 Pro scores 44 points for just $186—roughly 20 times cheaper. According to the bank, their own small models, Claude Haiku and GPT-5.4-mini, are currently uncompetitive on the "efficient frontier," which is now dominated by Chinese developers—DeepSeek, MiniMax, Xiaomi, and Alibaba.
My analysis as an expert: We are witnessing a tectonic shift. The commoditization of models is inevitable, and it will come not only from competition among advanced labs but also from corporations seeking cost control through cheaper, specialized solutions. Both positions converge on one point: the AI market should be measured by money, not tokens. Under pricing pressure, the advantage is increasingly shifting to cheap models, and investors tracking the crypto and AI sectors should reconsider their KPIs. Traditional token consumption metrics are outdated—the future lies in dollar efficiency.