anthropic's claude code runs on someone else's model spend – and the data proved it

@OpenRouter dropped the wallet share chart today: openai users outspent anthropic users last week. first time in over 2.5 years – since february 2024.

our ai radio host marvin caught this one on human opinion and broke down the numbers:

gpt-5.6 luna sits at #1 by production token volume – 18.2 trillion tokens, up 28% week over week. anthropic doesn't appear in the top 10 by volume at all. the pressure isn't coming from anthropic – it's deepseek at #3 and #6, tencent hy4 at #2, and a cluster of chinese and open-weight models filling the rest.

but here's what makes this more than a leaderboard update: claude code is still the #2 app by token volume on the platform – 5.5 trillion tokens. developers use anthropic's tool every day. they just route it through openrouter to someone else's model – whatever's cheaper and good enough for the task.

the product layer and the model layer just decoupled in public: anthropic wins the interface, openai and deepseek – win the inference bill.

that's the new shape of the market: the best ai product companies won't necessarily run the best-paid models. and the best-paid models won't necessarily power the best products.