Key Notes
- OpenAI reached just over half of combined OpenAI and Anthropic spending on OpenRouter in an early September week.
- Anthropic’s newer 3.6-fold increase in average token volume measures usage rather than spending or revenue.
- OpenRouter provides a view of one purchasing channel, excluding transactions made directly with AI developers or elsewhere.
OpenAI narrowed Anthropic’s lead in business spending through OpenRouter in September, underscoring a more competitive market for artificial intelligence models. The shift offers evidence of customer demand, although it measures a particular purchasing channel rather than the entire enterprise AI industry.
Newer figures show how quickly that picture can change. On October 25, OpenRouter’s Peter Walker said Anthropic’s seven-day average token volume had risen 3.6 times since the launch of Opus 5.5. That measures usage rather than dollars spent, so it cannot establish which company currently leads in spending.
The Wall Street Journal reported on October 7 that the companies roughly split September spending among about 120,000 businesses using both through OpenRouter. Anthropic had commanded around 75% of that spending in January. The monthly business sample and newer token figures describe different periods and metrics.
New from OpenRouter: 7-day average token volume for Anthropic models is up 3.6x since the launch of Opus 5.5
— Peter Walker (@PeterJ_Walker) October 25, 2026
September Spending Showed a Competitive Shift
OpenRouter’s own September research provides a separate weekly view. In the week beginning September 7, OpenAI accounted for just over 50% of combined spending on OpenAI and Anthropic models through the platform. It was the first week at or above that threshold since February 2024.
In the week beginning September 14, OpenAI also accounted for 78% of the two companies’ combined token volume. OpenRouter attributed its gains to a mix of a cheaper, high-volume model and a more expensive frontier offering. Its charts exclude reseller activity.
The denominator matters: those percentages compare the two suppliers with each other. They do not mean OpenAI captured half of all AI spending, or that other model developers disappeared from the competitive landscape.
A Platform Sample Is Not the Enterprise AI Market
OpenRouter gives developers a common way to access models from multiple providers. Its customers therefore offer a useful window into how buyers choose between competing products when they can change models within the same service.
However, transactions conducted directly with an AI developer or through another cloud platform fall outside that window. The data cannot capture every enterprise contract, employee subscription or privately negotiated deal across the industry.
It is also not enough to assume that all of the largest customers are absent. OpenRouter offers enterprise services. Without a fuller breakdown of the sample, its size distribution and purchasing patterns remain uncertain.
Customers using a model gateway may be particularly interested in testing alternatives or managing several providers. That is a reason to treat the results as a channel-specific signal, rather than extrapolating them into a comprehensive market-share estimate.
Why Companies Pay an Intermediary
The claim that buying through OpenRouter is always more expensive than buying directly also needs qualification. According to its pricing page, inference is billed at the provider’s list price without a markup on individual requests. OpenRouter instead charges when customers buy credits: 5.5% on its Standard plan and 8% on Business, with enterprise discounts available.
Those fees buy services beyond access to a single model, including routing, spending controls and a common interface for multiple suppliers. For a business, the relevant comparison includes the engineering and administrative costs of operating several separate integrations.
Direct contracts, discounts and promotions can change the comparison. Published list prices alone do not establish which route is cheapest for every customer or workload.
Token Growth Does Not Equal Revenue Growth
Tokens are units of model input and output. Spending depends on both how many are consumed and the prices charged, including the mix of models used. Requests measure something different again: a single request can involve a small prompt or a lengthy task.
As a simple illustration, doubling token consumption while halving the average price would leave spending unchanged. That is why Anthropic’s newer usage rebound cannot be treated as a matching increase in revenue, nor used to calculate an updated spending lead.
For buyers, lower prices can make more AI tasks affordable. For suppliers, the financial benefit depends on whether greater demand offsets lower prices and the cost of serving it. Platform traffic alone does not resolve that question.
AI Valuations Put Revenue Quality in Focus
The competition has financial significance as both developers seek capital. As we previously reported, OpenAI has been in talks about another major funding round. Anthropic’s potential listing has also drawn attention following its reported results.
Investors evaluating those businesses need to distinguish consumption growth from revenue, customer retention and cash generation. The OpenRouter figures demonstrate that spending choices can shift between major developers. They do not establish the profitability of that demand or settle which company has the stronger overall enterprise business.
