If you've narrowed your shortlist to two OpenAI-compatible gateways, you've probably landed on Kumo and OpenRouter. They start from the same place — one base URL in front of many models — so the real decision is about what each one optimizes for.
Same starting point: both are drop-in
Neither one asks you to rewrite anything. Both speak the OpenAI API for connected text endpoints, so you point your existing SDK at a new base URL, swap the key, and keep the rest of your chat/completions-style code. That also means the cost of trying either is one line, and the cost of leaving is the same one line back. No lock-in is the honest default here, on both sides.
Where the pricing comes from
This is the real difference. OpenRouter bills a per-model rate against credits — compare it directly on openrouter.ai. Kumo quotes an effective rate: the price after Kumo's automatic optimizations, shown for each model before you spend. Stacked, that lands 30–50% below list.
The honest catch: that range moves with your usage pattern and volume, so it isn't identical for every account. Either way you see the exact per-model rate before you spend a token — so you compare the two on numbers, not adjectives.
Privacy and spend controls
Kumo doesn't log prompt bodies — only the billing metadata needed to meter a call (token counts, model, timestamp). For anyone facing a security review, that's the line that matters; check OpenRouter's privacy policy for its own stance. On spend, both offer per-key limits; Kumo adds one prepaid balance across every model with per-key caps and alerts, so finance reconciles one bill.
When a broader public catalog matters
OpenRouter carries a much larger model catalog across far more providers, and it's public, established and proven at scale. If you're experimenting across dozens of models, that broader public catalog can matter more than Kumo's rate and billing model. Kumo is in private early access with a design-partner program and a curated catalog of flagships across OpenAI, Anthropic and Google — deliberately narrower, optimized for effective rate and team billing rather than breadth.
How to actually decide
Don't decide on a table — decide on your own traffic. Because both are drop-in, you can run the same workload through each for a week and read the real bill. If you need the widest possible catalog or the most established option, that points to OpenRouter. If you want a lower effective rate, one balance with spend controls, and prompt bodies that aren't logged, that's the Kumo case.