Developers have found a new favorite way to spend their compute budgets. According to OpenRouter data, Anthropic’s average daily token use rose about 3.5 times between September 22 and October 24.
The more telling detail is that spending rose too. Cheaper models often inflate usage numbers without much real money changing hands, and that does not appear to be the full story here.
What the OpenRouter numbers show
OpenRouter’s figures put the increase in Anthropic’s average daily token use at about 3.5x over the September 22 to October 24 window. A separate cut of the data, shared by Peter Walker, Head of Insights at OpenRouter, shows the 7-day average Anthropic token volume on the platform up 3.6x since the launch of Claude Opus 5.5 on September 22, 2026.
The two figures measure slightly different things. One tracks average daily use across a fixed window, while the other is a rolling weekly average pegged to the Opus launch. Both point in the same direction, and that direction is steeply up.
The model lineup behind the surge
Anthropic did not ship one model and wait. It released three in quick succession.
Opus 5.5 arrived first on September 22, 2026. Sonnet 5.5 followed on September 28, and Haiku 5.5 launched on October 7, 2026.
Each model sits at a different price point. Pricing is measured per million tokens, with separate rates for input (what you send the model) and output (what it sends back).
- Opus 5.5: $4 for input and $20 for output per million tokens
- Sonnet 5.5: $2 for input and $10 for output per million tokens
- Haiku 5.5: $0.10 for input and $0.50 for output per million tokens
Haiku is the eye-catcher. It is pitched as delivering performance close to Sonnet at a fraction of the cost.
Run the math and Haiku costs one-twentieth of Sonnet’s price on both input and output.
Why the spending data matters
Walker addressed that directly. He confirmed that expenditure on Anthropic’s models has risen alongside the token volume.
That points to genuine demand growth rather than a statistical side effect of cheap cached reads. People are not just sending more tokens. They are paying for more work.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.