AI Market Analysis: The Dollar, Not the Token, is the Key Indicator of Model Strength
Managing partner of venture capital firm Dragonfly, Haseeb Qureshi, made a significant statement that overturns the conventional understanding of evaluating the AI model market. In his opinion, token share is an extremely unreliable and distorting metric. Models should be compared solely based on the dollars spent on their usage. Analysis based on raw token consumption on the OpenRouter platform leads to systematic errors and incorrect conclusions.
Qureshi highlights four key problems with this approach. The first is massive subsidies. Chinese laboratories regularly launch new models with huge discounts or even free access. This attracts users who jump from one free model to another, artificially inflating token consumption without spending any real money.
The second problem is model size. Small models, such as Qwen 3.5-27B, cost about a hundred times less per token than the flagship Claude Opus. An increase in Qwen usage on the chart will look like a sharp jump in the share of open models, although in monetary terms it is an economically insignificant amount. According to him, analysis should be done within weight categories by model size.
The third problem is multi-agent systems. For the same amount of money, you can run a complex multi-agent architecture based on DeepSeek or GLM 5.2, which will "burn" several times more tokens than a single advanced model like Opus or GPT-5.5 Pro. The chart will show about an 18% loss in Opus's share, while actual spending will shift by only 5%. "Such charts exaggerate the importance of low-value tokens," Qureshi emphasizes.
The fourth problem is the OpenRouter platform itself. Large companies, once they settle on a single leading laboratory, prefer to contact Anthropic or OpenAI directly, bypassing OpenRouter's markup. On the chart, this looks like a decline in the US share, even though the tokens simply move outside the platform. Qureshi's conclusion: OpenRouter is useful for assessing the share within open models, but is not suitable for comparing open and closed ones.
A similar idea is developed by SageRoad Research founder Trevor Noren, linking it to pricing pressure on the industry. He cites a JPMorgan estimate, according to which future token consumption will primarily come not from advanced models, but from small open models sufficient for specific tasks.
According to JPMorgan, Amazon already offers about half a thousand open models at a price that is a fraction of the cost of advanced ones. Nvidia, together with Dell, Lenovo, and HP, is creating computers for AI agents. Meanwhile, their own small models, such as Claude Haiku and GPT-5.4-mini, are still uncompetitive on the "efficiency frontier," which is currently dominated by Chinese developers — DeepSeek, MiniMax, Xiaomi, and Alibaba.
The cost example is particularly illustrative. Running the Artificial Analysis Intelligence Index benchmark on Claude Opus 4.8 costs $3,700 with a result of 56 points, while DeepSeek V4 Pro scores 44 points for just $186 — roughly 20 times cheaper. The conclusion: advanced intelligence is not needed for everything, but only where it is necessary. GLM 5.2 from Z.ai appears comparable to the top models from Anthropic and OpenAI.
Noren believes that the commoditization of models will come not only from competition among leading laboratories but also from companies seeking cost control through cheaper, specialized models. In his assessment, corporate spending remains the most viable path for cloud giants to recoup their AI investments, but companies will spend as little as possible.
Expert opinion: Both positions converge on one point: the AI market should be measured by money, not tokens. Under pricing pressure, the advantage is increasingly shifting toward cheap models. This is a fundamental shift that investors and analysts must take into account when evaluating the sector. Those who continue to look at raw token consumption risk missing the real picture of the market.