Best API

Looks like a really useful service, especially for indie developers trying to keep costs under control. Nice that you can keep the same OpenAI-style workflow without rebuilding everything from scratch - that’s a big плюс. Simple idea, but very practical

useful product, looks cool

The "just change the base URL" approach is really smart. Most cost-optimization tools require you to rethink your entire architecture, but keeping OpenAI compatibility means zero friction for teams already in production. Curious how latency compares to direct OpenAI calls — that's usually the tradeoff with proxy layers.

The "just update base URL" claim is compelling for production systems, especially for minimizing friction. Alexandra correctly brings up latency; any proxy layer inevitably introduces some overhead. I'd add questions around potential model versioning issues. For larger projects, the security implications of proxying all API traffic are also a critical consideration.

Love to try, the llm will start to charge full price, awesome

The "just update your base URL" positioning is smart because it removes the biggest friction in API switching — developers already have mental overhead from the product they're building, and asking them to refactor API calls on top of that kills adoption. The real test for something like this is model parity and reliability; if outputs diverge subtly from direct OpenAI calls, it creates debugging nightmares downstream. Would be useful to know how you handle model versioning — when OpenAI ships gpt-4o-mini updates, does Best API track those automatically or pin to a snapshot?

Smart approach to reducing API costs without changing existing code. The proxy-based model means zero migration effort — that's a huge selling point for teams already in production.

This is a clever solution to a real problem. OpenAI API costs add up fast, especially if you are running multiple agents or processing large volumes of text. The drop-in replacement approach is key — nobody wants to refactor their codebase just to save on inference costs. How does the quality compare on more complex reasoning tasks? That is usually where cost-optimized routing gets tricky.