Slide 1
There's a new API that routes every LLM call to whichever provider runs it cheapest. It's called Auriko.
Change one line: your base URL
Here's the whole setup. Take your normal OpenAI client and change one thing: the base URL. Your code stays the same, and one key now reaches a dozen providers.
One request, a dozen prices
Why bother? The same request costs different amounts depending on where it runs. Auriko measured it: for one identical call, the priciest provider cost about four times the cheapest.
It reads the venues live
So how does it pick the cheapest? It watches the providers live and models each one's prompt caching. So cheapest means your effective cost, not the sticker price. You can also optimize for latency or throughput.
Their benchmark: about a third off
95% CI 30.6–34.9
Two things stand out. First, zero markup on your tokens: Auriko earns from a subscription, not a cut of your calls. Second, a published benchmark. On their own tests, routing cut cost about a third against a rival router, winning about eighty-five percent of calls.
A real platform, a sharper take
But there's a catch. Routing gateways already exist, so this is a late, sharper entrant. Its edge is cache-aware modeling and zero markup. And the savings depend on your model mix: big on open models many providers host, small on first-party frontier ones. So it's a real platform, not a feature. Just don't expect a third off everything.
Free tier, $89/mo after
There's a free tier with no markup, then eighty-nine dollars a month for unlimited use. Built by Michael Yang, who traded options in a past life and brought a quant's eye for the lowest price to your API bill.
