We show our math

How the router, caching, and sourcing cut your bill — measured, not marketed.

Our margin comes from buying tokens smarter than you would — not from a markup. That’s what funds the zero-fee, at-list pricing. Here is exactly how it works, and how we measure it. No “500× cheaper” claims; every number below is arithmetic you can check.

Three mechanisms

Complexity routing

Each request is scored. Simple turns go to a model that’s ~5× cheaper; hard turns stay on the full model. If a cheap answer looks incomplete, it escalates automatically.

Prompt caching

Coding sessions reuse a large, stable prefix. Cached input is billed at a fraction of fresh input, so a warm session costs far less than a cold one.

Multi-host sourcing

The same weights sell for different prices across hosts. We source from the cheapest host that passes our quality bench, and fail over if one degrades.

A worked example

Blended cost per 1M tokens for glm-5.2 at a 4:1 input:output mix (0.8 × input + 0.2 × output):

Scenario	Blended cost / 1M	vs. list
List sourcing, cold, no routing	$2.00	—
+ 50% prompt-cache hit	$1.54	−23%
+ 40% of turns routed to `glm-4.5-air`	$1.05	−47%
+ sourced from a discount host that passed the bench	~$0.63	−69%

You are billed at list; the gap between list and our actual cost is what the stack saves — and when a downgrade happens, you’re billed at the cheaper model’s rate, so part of that saving lands directly on your invoice.

The quality gate

Cheaper sourcing is worthless if the model gets quietly worse. Before any host or lane goes live it must pass a bench measuring tool-call fidelity, malformed-argument rate, latency, and output-cap behavior against a baseline. A host that regresses doesn’t enter rotation. This is why we can source cheaply without the “quantized junk” problem that plagues blind aggregation.

What we report

Your dashboard shows real, per-request numbers pulled from the metering ledger: tokens in/out/cached, which model answered, and the router’s counterfactual savings. Nothing here is estimated marketing math — if we can’t measure a saving, we don’t claim it.

See pricing Trust & data