Blog

OpenRouter vs Portkey: which layer are you actually buying?

These two get cross-shopped constantly and they are not the same product. One sells breadth, the other sells control, and the fee structures cross over at a number worth knowing.

By the Unbiased Team · published August 9, 2026 · fees and prices verified July 28, 2026; measured data dated in place

OpenRouter is an aggregator (400+ models, 5.5% on credits, breadth as the product). Portkey is a control plane ($49/month flat, tokens billed direct by your providers, governance as the product: SSO, audit logs, spend controls). Below ~$900/month of spend OpenRouter's fee is cheaper; above it, Portkey's flat fee wins on cost too.

Two products, one shelf

OpenRouter's product is the catalog: one API key, four hundred plus models, provider list prices passed through, and a 5.5% fee on the credits you load. Portkey's product is control: your traffic runs through a governance layer with SSO, audit logs, and spend controls for $49 a month flat on the Production tier, while tokens bill direct from your providers. There is also an open-source gateway edition if you want the software without the hosting.

They get compared because both sit between your code and the providers. But buying breadth and buying governance are different purchases, and mistaking one for the other is how teams end up disappointed in a perfectly good tool.

The fee math

The structures cross over just under $900 a month of model spend. Below it the 5.5% is the cheaper convenience; above it the percentage compounds while the flat fee does not: at $5,000 a month, $3,300 a year against $588. Run your own number in the fee calculator or the explorer below.

One nuance favors Portkey at scale beyond the fee itself: tokens billed direct by your providers means your negotiated provider terms, if you have them, stay yours.

The question neither answers

Both leave model selection with you: a dropdown on OpenRouter, routing rules you write on Portkey. Selection is where the real money moves; on our measured four-task workload, right-model routing saved 14x what any fee cost. Both tools assume you will do that homework, and the market reassigns it constantly: three flagship-tier price events in the last nine months.

If the homework is the part you want gone, that is the blended-model category: Pareto runs several LLMs on every request, synthesizes one answer, and bills at cost. We publish receipts on the model card; we are also a competitor to both, so verify.

Questions, answered