Blog

What is OpenRouter and how does it work?

One API key, four hundred plus models, provider list prices passed through untouched. Here is what OpenRouter actually is, what the fee structure costs, and the two questions the catalog cannot answer for you.

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

OpenRouter is an LLM aggregator: your code speaks one OpenAI-compatible API and OpenRouter routes requests to 400+ models across many providers at their list prices. It charges a 5.5% fee on credit purchases (BYOK traffic gets 1M requests a month free, then 5%). You still choose the model; OpenRouter handles the integration.

What it does

Every LLM gateway solves the same integration problem: one API surface in your code, many providers behind it. OpenRouter is the aggregator version of that answer. You load credits, point your OpenAI-compatible client at their endpoint, and pick any model in a catalog of four hundred plus, from frontier flagships to open-weights served by a dozen competing hosts.

Token prices pass through at provider list rates. The fee sits one level up: 5.5% on every non-crypto credit purchase (minimum $0.80), 5% flat for crypto, and bring-your-own-key traffic gets its first million requests each month free, then a 5% usage fee. No subscription, no per-seat cost.

What it costs at your scale

A 5.5% fee is small at prototype scale and a real line item later. The structural crossover sits just under $900 a month of spend: below it, the percentage is cheaper than a $49 flat-fee control plane; above it, the percentage becomes a growing tax. At $5,000 a month the fee runs about $3,300 a year, 5.6x the flat alternative.

Keep the fee in proportion, though. On our measured four-task workload, routing the work to the right models saved 14x what the platform fee cost. The fee is not the money; the model selection is the money.

The two questions the catalog leaves open

First: which model. The catalog puts the whole market one dropdown away, but the dropdown is yours to operate, and the market moves. The flagship tier repriced 67% overnight in November 2025, Sonnet 5 has a scheduled 50% increase on September 1, and new models land monthly. Every event reopens your choice.

Second: which provider. The same open-weights model served by different hosts returns measurably different quality; OpenRouter built telemetry across billions of tool calls and launched Exacto because tool-call accuracy varies by provider. Same name and same price on a catalog does not mean same output.

Where it fits, honestly

If you are prototyping across many models, an aggregator is the right tool and OpenRouter is the biggest one; breadth is the product and the fee is cheap at that stage. If your constraint is governance, a control plane like Portkey fits better. If data cannot leave your infrastructure, self-host LiteLLM.

And if what you actually want is the bill down without making model selection your standing job, that is a different category: a blended model like Pareto runs several LLMs on every request, synthesizes one answer, and bills at cost, with receipts published on the model card. We are a competitor; check our numbers the way you would check theirs.

Questions, answered