Crypto news

Recently, a discussion has been gaining momentum in the artificial intelligence industry about how traditional metrics based on token consumption provide a distorted picture of market dynamics. Leading analysts and managing partners of major venture capital funds agree: for an objective assessment of the AI model market, it is necessary to focus on cash flows, rather than raw token usage indicators.

Why Are Tokens an Unreliable Indicator?

The main problem lies in the pricing structure and market incentives. Chinese laboratories regularly release new models to the market with aggressive discounts or even free access. This naturally attracts users who migrate between free solutions, artificially inflating token consumption volumes but generating no real monetary revenue. This approach creates an illusion of high popularity that is not backed by economic value.

The second fundamental problem is the difference in model sizes. Small models, such as Qwen 3.5-27B, can cost about a hundred times less per token than flagship solutions like Claude Opus. A sharp jump in the usage share of open models on charts may look like a tectonic shift, although in economic terms it is an insignificant amount. Correct analysis requires comparison within weight categories by model size.

The third problem is related to the emergence of multi-agent systems. The same amount of money can be spent either on a complex architecture based on DeepSeek or on a single advanced model. With comparable performance, a multi-agent configuration will "burn" significantly more tokens for the same money. As experts note, if 5% of Opus usage migrates to such a system with a fourfold token consumption, the chart will show an 18% loss of Opus share, although actual spending will decrease by only 5%. This leads to a dramatic exaggeration of the significance of cheap tokens.

Finally, the choice of analysis platform, such as OpenRouter, also introduces distortions. Large companies, having settled on one provider, prefer to work directly with Anthropic or OpenAI, bypassing the aggregator's markup. On charts, this looks like a decline in the share of American models, although the tokens simply leave the platform. OpenRouter is useful for assessing the share within the segment of open models but is absolutely unsuitable for comparing open and closed solutions.

Is the Future in Cheap Models?

Price pressure on the industry is only intensifying. According to JPMorgan estimates, 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 specifically for AI agents. Meanwhile, their own small models, Claude Haiku and GPT-5.4-mini, are still uncompetitive on the "efficient frontier," where Chinese developers—DeepSeek, MiniMax, Xiaomi, and Alibaba—dominate.

The figures are particularly telling: running a set of tasks on Claude Opus 4.8 costs $3,700 with a result of 56 points, while DeepSeek V4 Pro scores 44 points for just $186—about 20 times cheaper. The conclusion is obvious: advanced intelligence is not needed everywhere, and where it is necessary, Chinese models are already comparable to the top solutions from Anthropic and OpenAI.

The commoditization of models will come not only from competition among leading laboratories but also from corporations seeking to control costs through cheaper, highly specialized solutions. Corporate spending remains the most viable path for cloud giants to recoup their AI investments, but companies will minimize costs.

Expert opinion: The AI market is entering a phase where the winner will not be the one with the smartest model, but the one who offers the best price-to-quality ratio for a specific task. Investors and analysts should shift their focus from token consumption metrics to analyzing real cash flows and margins. It is the dollar, not the token, that will become the main measure of success in the new era of artificial intelligence.