Unit-based versus value-based pricing
The economic climate around artificial intelligence is moving incredibly fast, but one of the most fascinating dynamics to watch right now is how LLM models are actually priced. It represents a fundamental tension between two different economic frameworks: unit-based pricing and value-based pricing.
How this tension resolves will dictate how these models deliver value, and ultimately, how the frontier labs themselves are valued by the market.
Defining the Two Pricing Worlds
To understand where AI is heading, it helps to look at how we price things in the traditional economy.
Value-Based Pricing: This happens when a product or service is charged for based on how much it is worth to the end user, completely decoupled from the cost of production. Take a high-end consulting service as an example. If a consultant can save a company $100,000, it does not really matter what it cost the consulting firm to put the deck together. Even if it only cost them $45,000 to build, they can charge based on the massive utility delivered and capture the profit. Another great example is a bespoke service like Stripe’s fraud prevention, where the price is tied directly to the value and security created for the consumer.
Unit-Based Pricing: This is the world of commodities, like electricity or oil. When there are many different alternatives that do the exact same thing, you cannot charge a premium based on abstract worth. Instead, the market forces you to charge the lowest price possible: the bare cost of production plus a very small margin.
The LLM Token Dilemma
When you look at LLM tokens, there is an ongoing argument across the industry, especially from the frontier labs, about where on this spectrum AI should live. Right now, the market is split into two distinct worldviews.
World 1: The Value-Based Premium
In this world, the tokens generated by frontier models carry a distinct premium. They provide a unique, value-based advantage to the end customer that simply cannot be replicated by other companies. Because the model can solve complex, high-value problems that cheaper models cannot touch, the labs can price their API access based on the massive ROI they unlock for businesses.
World 2: The Unit-Based Commodity
The alternative world is driven by the rise of open-source and highly capable alternative models. If a cheaper model can provide basically the same value for a fraction of the cost, the premium evaporates. The moment intelligence becomes a repeatable commodity, the economic model aggressively compresses back toward unit-based pricing.
The Shift Toward Compression
What we are seeing play out right now is a noticeable compression back towards unit-based economics.
Even with the latest frontier models, developers are finding that there are clever trade-offs to be made in how these tools are deployed. You can achieve excellent results and maintain high token throughput without relying exclusively on the most expensive models.
This shift is highlighted by the rise of modern coding harnesses and multimodal routing frameworks. These systems give companies a massive advantage because they can essentially delegate work on the fly. Instead of burning expensive frontier tokens on basic tasks, the harness routes simpler queries to cheaper, highly efficient models, only calling in the premium models when absolutely necessary.
Where Do We Go From Here?
It is still too early to tell where the equilibrium will settle. However, this shifting dynamic tells us a lot about the current economic climate surrounding AI. Watching whether LLM tokens can maintain a value-based premium or if they will succumb to the race-to-the-bottom world of unit-based commodity pricing is easily one of the most compelling pieces of the tech landscape to watch right now.