🏁 I animated the token 💰 spend race, from ~lifetime Vercel AI Gateway usage, which aggregates trillions of tokens from millions of developers a month. Fascinating to see the fluctuations among the labs, Anthropic's dominance, and the rise of open weight AI.

I’m suprised by Google tbh

Google is a quiet killer, because their models make economic sense to people. Even if @x is not excited about them as much (or at least in my bubble)

@X We are in a similar bubble. My entire timeline isn’t excited at all. Hell neither am I. Not sure what I’m missing here other than from and economic standpoint. IMO OpenAI is the better route from and economic stand point balanced by actual quality.

would be funny to add events like "Anthropic outage" and see people get auto rerouted to OpenAI or whatever :D that's the whole point of AI gateway right

We have several token origins for Anthropic's models, so in practice it's almost never offline. (And we recently shipped routing rules in case a model gets retired!)

tunes mixed by dj G

By @suno actually 😂 it's awesome

Interesting that you chose token spend. Just because Anthropic charges an arm and a leg doesn’t mean it’s good. This is the graph of token volume. People are doing a lot more work with other models.

So confused as to why people are using Google models…

@dhruv_rathee ye bhi copy krle

This is in number of tokens or dollar spent ?

Who was saying Google models are not used anymore 😁

google openai fighting for their life

Why is Anthropic so far ahead?

@naaavinash

sick

Am curious, which open weigh model is being used most?

can you share by modality as well

bragging about how many tokens your agent chewed through is the 2026 version of bragging about your AWS bill 😆 super interesting graphic though

Who is using google? I thought it would have been OpenAI and Anthropic

what % do you estimate Open Weight will be at the end of 2026?

The bump/fall is like crypto between providers

Anthropic's dominance is just devs burning tokens trying to fix a broken Webpack config 🙃

You look like the type to enjoy listening to an alt-right podcast as you furiously jerk off to anime hentai, feeling guilty for being an edgelord but too addicted to all of it to quit.

@rauchg token numbers or $$$?

anthropic dominating makes perfect sense. it’s literally just the collective token burn of every developer yelling at cursor to fix a broken build at 2 AM.

Anthropic dominates token spend. OpenAI and Google tied. Open weight at 0.7%. Token spend is the real benchmark—and the market is voting with its wallet.

Guille, el martes juega Argentina. Cámbiame el horario del workshop de eve, please 🙏

It would be interesting to see similar stats from Cursor team. Composer 2.5 is a great cost-efficient daily driver and I sometimes wonder how many people know about it.

It would be great if you could publish the categories by company.

You mean of the consumers of tokens?

I mean a breakdown of the customers by industry for each AI provider - which kinds of companies are using Anthropic vs. OpenAI vs. Google? very curios about the Google one.

Which models are being used the most? Any surprises in the top 3?

man, 1.6% for open weight is rough. all this open source comeback talk and i'm just seeing a rounding error. developers vote with their wallets, not their GitHub stars.

To work with Google model you either use openrouter or Vercel, because the native API key and budget management in Google is poorly made.

Interesting

Anthropic earned it. Claude Code and Cursor are the only reasons my token bill is scary right now. 💸

Just played around with Gemini and oh boy... how bad is this model - compared to any of Claude's models? It's like GPT 3 vs 5.

Did not expect Anthropic to dominate like that. Interesting. 🤔

代币消耗比单纯看模型热度更接近真实开发需求。实验室份额波动说明应用层在快速试错，开源模型的存在感也会继续增强。

wow google api use is so much higher than i thought. the fact that they left off ahead of openai is wild to me. was also certain that OAI retook the lead after May 2026. x is really a bubble