This graphic from a16z, sourced from YipitData, tracks the token consumption of Claude Code integrations on OpenRouter from late December 2025 to mid-February 2026.
During the week of December 22, the token consumption was zero. Nothing at all.
By the week of January 5, it had just begun to emerge, accounting for less than 0.5%. Claude Code generated a small amount of tokens through Anthropic and third-party channels.
The week of January 12 saw a slight increase to about 1.5%.
From January 19 to January 26, it hovered around 0.8%, even dipping a bit. It seemed like a lukewarm new product.
Then came OpenClaw.
In the week of February 2, token consumption jumped to around 6.7%. That deep red bar represents OpenClaw's token usage, transforming from zero to a major player in just one week.
By the week of February 9, it soared to 13%. It doubled! OpenClaw’s token count skyrocketed to about 180 billion, surpassing the total from all previous weeks combined.
In contrast, the token consumption of Claude Code through Anthropic and other channels (the green and blue bars) remained relatively stable, around 10-20 billion. This means that this explosive growth was almost entirely driven by OpenClaw.
What does this indicate?
The demand for AI programming tools is real and immense. OpenClaw's ability to capture 13% of traffic on a mainstream routing platform in just six weeks shows that developers' needs for AI code generation are far from being met.
Open-source and third-party integrations are becoming the primary distribution channels. While token consumption for Claude Code through official Anthropic channels hasn’t changed much, it exploded through third-party integrations like OpenClaw. The ecosystem's power outweighs that of official channels.
The growth curve is accelerating. It only took a week to move from 6.7% to 13%, and there are no signs of slowing down. If this trend continues, OpenClaw could soon command 20% or more of OpenRouter's total.
A third-party integration tool achieving 13% of a platform's total in just six weeks? The pace of the AI programming landscape is accelerating rapidly, leaving little time to react!
Top comments (0)