9 of the world's top 10 open-source LLMs are now Chinese. After GLM-5.2 landed, the
only non-Chinese model still in the top 10 is Llama. If your gateway taxes every call
by 5.5%, you are paying that tax to route to models you could reach for free.
The 5.5% you didn't notice you were paying
OpenRouter's pitch is fair: one key, 400+ models, transparent pricing.
But read the billing page closely. Every credit-card top-up adds 5.5% (minimum $0.80). Crypto top-ups add 5%. Token prices look "at cost" — until you check community benchmarks and notice DeepSeek-R1 routinely runs ~15% above what direct providers charge, depending on which underlying provider OpenRouter routes to that hour.
For a hobby project, this is invisible. For anyone running a real workload, the math gets uncomfortable fast.
$10,000 / month on inference -> ~$550 / month in routing tax
$100,000 / month -> ~$5,500 / month
$1,000,000 / month -> ~$55,000 / month
That's not a rounding error. That's an engineer's salary. And you're paying it to do something you could do with a Cloudflare Worker and ten lines of Go.
I built , top up nothing, and you get $1 in free trial credit — enough to hit DeepSeek-R1, Kimi K2, Qwen3, and GLM-4.6 a few hundred times each from the same key. No card. PayPal works. The base URL is https://api.haotokai.com/v1, the SDK is whatever you already use, and the migration from OpenRouter is the diff above.
If you stick around after the free credit, great. If you don't, you've at least seen what the 5.5% was actually buying you — and you'll never look at an aggregator's billing page the same way again.
That alone, I'd argue, is worth ten minutes.
SOCIAL SHARE CARD GENERATOR