Join the waitlist

“Cheapest LLM API” usually smells like a scam. Here’s one with receipts.

Most cheap-API pitches ask you to trust a discount you can’t inspect — resold quota, mystery keys. Ours is the opposite: keep your provider and your keys. The savings come from grading the work — the cheaper model only wins if it matches the quality — and every claim traces to a public run.

Join the waitlist Free · join before July 15, get $50 in credits when we open

The smell test

You’re right to be suspicious.

Discounted tokens have to come from somewhere. When nobody will tell you where, it’s usually resold quota or somebody else’s keys — and you find out the day they get cut off.

The usual “cheap” pitch

A discount you can’t inspect.

  • Mystery keys or resold quota sit between you and the model.
  • Your traffic runs through someone else’s account.
  • No way to check which model actually answered.
  • The audit trail is “trust us.”
This one

Cheap you can check.

  • Keep your provider. Bring your own keys. One OpenAI-compatible endpoint — a faithful pass-through: same models, same outputs, read-only.
  • The savings come from grading the work — a cheaper model only moves if it holds the quality — not from a mystery reseller.
  • Every claim traces to a public run, re-gradable from the transcripts.
  • $0 until you flip Pareto to production — a KEEP verdict pays us nothing.

And don’t expect your vendor to point out a cheaper grade: the cost-optimization docs at OpenAI, Anthropic and Google — 3 of 3 — never mention a competitor exists. Every vendor’s answer to “spend less” is “spend less with us.”

The receipt · Run №000 · 07.02.26

The cheapest 8/8 pass cost $0.0104 — 1/39th of the priciest.

Nine models. Identical prompt, one shot each, graded by an 8-point Playwright audit — not by taste. Five passed the same quality bar. The cheapest that did cost 1/39th of the priciest.

$0.0104
cheapest 8/8 pass
$0.4085
priciest 8/8 pass
Grok 4.3
8/8 · 41s
$0.0104
DeepSeek V4 Pro
8/8 · 244s
$0.0199
GLM 5.2
8/8 · 240s
$0.1210
Gemini 3.1 Pro
8/8 · 100s
$0.1440
GPT 5.5
8/8 · 128s
$0.4085

A five-way functional tie at a 39× price spread. That’s what “cheapest” means here: the lowest price that still passes the grade — not a discount, a verdict.

GPT 5.5 did ship the best-designed page — at 39× the price. When the frontier wins, the report says so.

Four models DNF’d on transport errors — failures in the pipes, not the models — receipts logged. Blind contests №001–005 are live on the public ledger ↗

The mechanism

Proof, not promises.

Step 1
Bring your keys.

Point your OpenAI and Anthropic keys at our pass-through proxy. Same models, same outputs, read-only. You keep your provider.

Step 2
Pareto — our model — runs in shadow.

It re-runs a sample of your real requests async, off the production path. Users never see it.

Step 3
You get the verdicts.

Quality and price per workload, re-gradable from the transcripts. Nothing moves without your sign-off.

MOVE verified savings — flip when ready
SHADOW still comparing — free
KEEP the frontier won — pays us $0

Pareto wasn’t ready to enter Run №000 — once it ships, it enters every contest under the same audit.

Our pricing is our proof. We charge nothing until you flip.
Proxy · shadow · reports$0
MOVE — Pareto in production≈ half frontier list
A KEEP verdict pays us zero — if the cheaper model can’t hold the quality, we make nothing. That’s why you can trust the MOVEs.

Live July 15

Send us your AI bill.
We’ll show you the difference.

We couldn’t confirm your signup from here, so we opened a pre-filled email in your mail app instead — hit send and you’re on the list. Nothing opened? Write to hello@unbiased.ai.
You’re on the list.

We’ll write when unbiased.ai opens on July 15. First shadow runs go to the list in order.

Join before July 15 → $50 in credits when we open · keys encrypted, never logged · free until you flip Pareto to production

Prefer email? Send your bill to hello@unbiased.ai