Vercel's AI Gateway data shows a clear split between AI usage and AI spending. Anthropic does not lead by token volume, but it dominates the money flowing through the platform.
The figures point to a market where cheaper model tiers are gaining ground overall, while premium models still command meaningful budgets from developers and teams that see enough value to keep paying.
Anthropic Takes A Bigger Share Of Spend Than Usage
According to the source data, Anthropic accounts for 65.1 percent of spending on Vercel's AI Gateway. Its token share is much smaller at 30 percent.
That difference matters because tokens are the basic unit of AI model usage. A provider that represents a relatively modest share of tokens but a much larger share of spending is earning more money for each unit of use than competitors on the same gateway.
The source states that Anthropic's per-token cost runs 4.4 times the average of every other provider. In plain terms, developers using Anthropic through Vercel's AI Gateway are paying substantially more per token than they are for other providers available through the platform.
This does not show the whole AI market. It shows what is happening inside Vercel's AI Gateway, which is only one channel. Still, the numbers are useful because they capture real developer spending behavior rather than only public discussion about model pricing.
Fable 5 Quickly Became A Major Spending Driver
Fable 5 appears to have gained commercial traction quickly on the gateway. It climbed to 13.2 percent of Gateway spending, making it second only to Opus 4.8 in that measure.
That is notable because Fable 5 is not described as simply taking usage from existing teams. The source says nine out of ten Fable teams in July were new customers.
Vercel reads that as a sign of two related trends:
- Fable 5 is bringing in fresh users rather than only shifting existing usage.
- Developers and teams still show willingness to pay for premium models.
This is the key commercial signal. If a model is expensive but still attracts new customers, the market is not only optimizing for the lowest possible token price. Some developers appear to be choosing higher-cost models because they believe the tradeoff is worth it.
Ramp Payments Saw A Different Slice Of The Market
The source also notes that payments company Ramp Payments had previously pointed to relatively low Fable usage as evidence that the model might be priced too high.
That view is not necessarily in conflict with Vercel's data. The article makes clear that both companies only see a slice of the full picture. A payments company and a developer platform may observe different customer groups, different workflows, and different spending patterns.
The contrast is still important. It shows why AI pricing signals can be difficult to interpret from a single dataset. Low usage in one view may suggest price resistance, while strong spending growth in another may suggest premium demand.
For developers, the practical lesson is that token volume alone does not explain model adoption. Spending share, new customer share, and relative price all change the meaning of the same usage data.
Cheaper Tiers Are Growing, But Premium Demand Remains
Across Vercel's AI Gateway, token volume rose 59 percent overall, while spending grew 37 percent. That means usage increased faster than spending.
The source also says average token prices fell 13.6 percent as companies shift toward cheaper tiers. This points to a broader pattern: more AI work is being routed through lower-cost options, reducing the average price paid per token.
At the same time, Anthropic's position shows that premium pricing has not disappeared. Even with cheaper tiers pulling down average token prices, Anthropic still accounts for a far larger share of spending than token use.
That combination gives a more nuanced picture of AI model economics. The market is not moving in only one direction. Some companies are clearly trying to lower average costs, while others continue to allocate budget to higher-priced models when those models fit their needs.
What The Numbers Say About AI Buying Behavior
The Vercel data suggests that AI buyers are becoming more segmented. Some usage is moving toward cheaper tiers, which lowers average token prices. But premium models can still attract new teams and capture a large share of spending.
For Anthropic, the numbers show strong pricing power on Vercel's AI Gateway. For competitors, they show the challenge of converting token volume into revenue. For developers, they show that the real cost of an AI stack depends not just on how many tokens are used, but on which provider handles them.
The most important takeaway is simple: lower-cost AI is growing, but expensive models are still being paid for. On Vercel's AI Gateway, Anthropic is the clearest example of that split.