Rauch says Anthropic fell from 69% to 40% of Vercel spend
Two months of gateway spend data shows developer dollars moving fast, and open models taking about half of the difference.
Guillermo Rauch posted the spend split across OpenAI, Anthropic and open models on Vercel AI Gateway for the last two months. Anthropic is still number one in spend, but it went from 69% to 40%. OpenAI went from 10% to 24%, and Rauch says OpenAI now leads in tokens.
Kimi K3 + DeepSeek took ~half of Anthropic's loss
Opus 5.5 is up to 10% of spend in 2 days
The part worth sitting with is where the rest went. Rauch says Kimi K3 and DeepSeek took roughly half of Anthropic's loss, which means the shift is not a simple two horse race between the big labs. He also notes GPT-6 Astra and GPT 5.6 Sol are driving OpenAI's climb, and that OpenAI accounts for 62% of image generations on the gateway.
The speed is the story
Rauch says Opus 5.5 reached up to 10% of spend in two days. That is the number that should change how you plan. A model ships, and within 48 hours a tenth of the money flowing through one of the larger developer gateways is going to it. Nobody is sitting on a twelve month procurement cycle here. Developers point their code at the new endpoint and the spend follows the same week.
This lines up with what Rauch reported earlier about open models taking the majority of gateway tokens. Tokens and spend are different measures, and the gap between them is the whole economics of this. Cheap models move a lot of tokens without moving much money. So a drop in Anthropic's spend share does not mean fewer people are using Anthropic, it can also mean everyone else got cheaper at the same volume.
What this is and is not
This is one platform's mix, self reported by the person who runs it. Vercel AI Gateway skews toward web developers building on Vercel's stack, which is not the same population as enterprise API customers on annual contracts. Treat it as a fast moving sample rather than the market.
What it is good for is the direction. If you are a solo founder budgeting inference for the next quarter, the read is that no vendor's share is safe for two months at a stretch, and that the cheaper open models are not a fringe experiment anymore. They are taking real dollars off the leader. The practical version of that is keeping your model calls behind something you can swap, because the price you negotiated in your head in January will look silly by March.