unbiased.ai
Contest #002Ran 2026-07-04
Contest #002 · One inbox · 25 tickets · Four leading models

The $0.03 inbox vs the $0.17 inbox —
and not a single ticket apart.

Four AIs worked the same 25-ticket support queue for CloudCanvas — refunds, a legal threat, a GDPR deletion, replies owed in Spanish, German and Japanese. Every one cleared all 25. We stripped the labels. Below is the hardest ticket in the stack: a legal threat that must be escalated, never argued. Read the four replies, guess which run cost 5.7× the most, then reveal the receipts.

TICKET #9 · legal · priority: highSpotlight: the legal threat
From: a customer's attorney — "My client has suffered damages due to your platform's downtime and we are prepared to pursue legal action. I expect a full accounting and compensation. Respond within 48 hours." Policy: acknowledge receipt, escalate to legal@cloudcanvas.com, do not discuss the substance, admit no fault.
AReply A
We’ve received your message. Because it includes a legal claim and references potential legal action, we have escalated it to legal@cloudcanvas.com. We won’t discuss the substance of the claim in this support thread. Regards, CloudCanvas Support
$•.••••
BReply B
Hello, Thank you for your message. We confirm it has been received, and because it references potential legal action, it has been escalated to our legal team at legal@cloudcanvas.com. They will handle all further communication on this matter. We are not able to discuss the substance of the claim through this support channel. Regards, CloudCanvas Support
$•.••••
CReply C
Hello, Thank you for your message. We have received your communication and have escalated it to our legal team at legal@cloudcanvas.com for their review. They will follow up with you directly. We are not able to discuss the substance of your claim through this support channel. Best regards, CloudCanvas Support
$•.••••
DReply D
Thank you for your message. We have received it, and because it references potential legal action and damages, I’ve escalated it to our legal team at legal@cloudcanvas.com, who will follow up with you directly. Out of respect for that process, I won’t discuss the substance of the claim here. Any further correspondence on this matter should go to legal@cloudcanvas.com.
$•.••••
All four escalate cleanly and admit no fault — a textbook pass. Guess which run cost the most, then reveal.
$0.0303
Cheapest run — the whole 25-ticket inbox
$0.1719
Priciest run — same 25 tickets
5.7×
Price gap, for a four-way tie on quality
25 / 25
Tickets cleared — by every model in the field

A four-way tie on quality. A 5.7× spread on the bill.

Every model cleared all 25 tickets — the refund denials, the legal escalation, the GDPR deletion, the German double-charge, replies owed in three languages. Nobody fabricated an invoice. Nobody caved on policy.

So the answer isn't in the replies — you can't tell them apart. It's on the invoice. One inbox cost six cents on the cheapest route and seventeen on the priciest, for work a grader scored identically. Most teams are hard-coded to the expensive end and never notice.

That's what Pareto fixes. It routes every ticket to the cheapest model that clears your quality bar, and hands you one endpoint you never think about. You keep the result and pay the low number.
A support desk at real volume
Hard-coded to the priciest model$687
Routed to the cheapest that clears the bar$121
You'd save — for replies a grader can't tell apart
$566/mo
$6,797 a year
Per-ticket cost (run ÷ 25) × monthly volume. Illustrative — one desk, one prompt; your real mix varies. Measuring that mix is exactly what the free audit does.
The scoreboard

Tickets cleared, out of 25

A ticket clears only if every criterion is met — right policy call, no fabricated facts, correct language, no forbidden exclamation marks. Four bars, all the way to the wall.

AReply A25/25
BReply B25/25
CReply C25/25
DReply D25/25
Perfect four-way tie. The traps caught no one: T2 (outside-window refund) denied without inventing an exception, T24 (usage overage) held as non-refundable, T7 (German double-charge) investigated rather than fabricated, T14 (GDPR) confirmed deletion within 30 days with a DPA reference.
Two more spotlight tickets

Same traps. Same blind labels. Read them side by side.

Ticket #2 is a refund demand that policy says to deny — the tempting wrong answer is to invent an exception. Ticket #6 must be answered in Spanish with the exact export path. Switch tabs; the labels stay withheld.

Ticket #2 — "I just noticed I've been charged for months. I want a full refund of everything." Team plan, billed monthly, first charge ~4 months ago. Correct call: deny — outside the 30-day monthly window — without inventing an exception, and admit no fault.
A$0.0820
Hi Priya, Thanks for reaching out. Self-serve monthly plans are eligible for a full refund within 30 days of each charge. Charges from about 4 months ago are outside that refund window, so we can’t refund all past charges. If you have a recent monthly charge from within the last 30 days, we can review that for refund eligibility and help cancel the subscription so there are no further renewals. Best, CloudCanvas Support
B$0.1719
Hi Priya, Thanks for writing in, and I'm sorry for the surprise on your statement. Our refund policy for self-serve monthly plans covers charges made within the last 30 days. That means the earlier months are outside the refund window and I'm not able to refund those. However, if your most recent monthly charge was within the last 30 days, that one is eligible — just confirm and I'll process a full refund of that charge right away. I can also cancel the subscription now so no further charges occur. Would you like me to do that? Best, CloudCanvas Support
C$0.0303
Hi Priya, Thank you for reaching out, and I understand the frustration of realizing you've been subscribed longer than intended. Our refund policy for self-serve monthly plans covers a full refund within 30 days of the charge. Since your subscription began approximately four months ago, the charges fall outside that window, and I'm not able to issue a refund. I'd be happy to cancel your Team plan so you aren't charged again going forward. You'll retain access through the end of your current billing period. Just let me know if you'd like me to proceed. Best, CloudCanvas Support
D$0.0939
Hi Priya, thanks for writing in, and I understand the frustration of a charge you'd lost track of. Here's where things stand: our refund window for monthly plans is 30 days from each charge. Since your first charge was about four months ago, I'm not able to refund the full history of charges. I can refund the most recent charge if it falls within the last 30 days, and I'll cancel the subscription right away so you aren't billed again going forward. Just confirm you'd like me to cancel and I'll take care of it. I'm sorry for the inconvenience here.
Ticket #6 — "¿Cómo exporto un tablero a PDF? Llevo dos días intentándolo." Correct call: answer in Spanish; export path is Share (top-right) › Export › choose PDF; promise no timeline.
A$0.0820
Hola Lucía, Claro. Para exportar un tablero: 1. Abre el tablero. 2. Haz clic en “Share” en la esquina superior derecha. 3. Selecciona “Export”. 4. Elige el formato PDF. También puedes exportar en PNG o SVG desde el mismo menú. Saludos, Soporte de CloudCanvas
B$0.1719
Hola Lucía: Gracias por escribirnos, y lamento la frustración de estos dos días. Exportar un tablero a PDF es sencillo: abre el tablero, haz clic en "Share" (Compartir) en la esquina superior derecha, elige "Export" (Exportar) y selecciona PDF. También puedes exportar en PNG o SVG desde el mismo menú. Si no ves la opción o algo falla al exportar, avísanos y lo revisamos contigo. Un saludo, Soporte de CloudCanvas
C$0.0303
Hola Lucía, Para exportar un tablero, abre el tablero que deseas exportar, haz clic en el botón "Share" en la esquina superior derecha, elige "Export" y luego selecciona el formato que prefieras: PNG, SVG o PDF. Si tienes alguna otra pregunta, no dudes en contactarnos. Saludos cordiales, El equipo de CloudCanvas
D$0.0939
Hola Lucía, gracias por escribir. Con gusto te ayudo a exportar tu tablero a PDF. Los pasos son: abre el tablero, haz clic en Compartir (Share) en la parte superior derecha, elige Exportar (Export) y luego selecciona el formato PDF (también puedes elegir PNG o SVG). Con eso deberías poder descargar tu tablero sin problema. Si la opción sigue sin aparecer después de abrir Compartir, avísame y lo revisamos contigo. Un saludo.

The verdict

All four cleared the inbox. Every refund call correct, every policy trap held, three languages handled, no fabricated invoices, no forbidden exclamation marks. On the grader it's a 25/25 four-way tie — which is exactly why we hid the labels.

So the only thing that ranks them is the bill, and the bill ranges 5.7×. Four inboxes you genuinely can't rank by quality, priced from three cents to seventeen.

Pick by brand and you're gambling with your budget. Pick the cheapest that still clears the bar — which is all routing is — and you keep the result while cutting the bill to a fraction. "We'll send you to whatever's cheapest that works" is a sentence no single lab can say. We can.

The receipts

All four cleared 25/25 — so this is a pure price race. One full 25-ticket run, one shot each.

ReplyCost / inboxvs cheapest
Reply CCHEAPEST$0.0303
Reply A$0.08202.7×
Reply D$0.09393.1×
Reply BPRICIEST$0.17195.7×

Four leading models ran this inbox — Claude Fable 5, Claude Opus 4.8, GPT-5.5, and one more. We're not telling you which is which, because you couldn't tell either. That's the point.

The answer

The answer isn't picking a model. It's routing to the cheapest one that clears the bar.

Every model in this contest cleared all 25 tickets — including the one that did it at 1/5.7 the cost of the priciest. Lock yourself to a single frontier vendor and you pay the top of that range on every ticket, forever. Pareto sends each request to the cheapest model that still passes your quality bar, proves it against the grader, and gives you one endpoint. You stop choosing. You just stop overpaying.

One shot each

Identical 25-ticket inbox, same system prompt, no cherry-picking. Each reply is exactly what the model returned.

Graded on the rules

A ticket clears only if every criterion is met — right policy call, correct language, no fabricated facts, no forbidden exclamation marks. All four cleared 25/25.

Real bills

Fable/Opus: delivered output chars ÷ 4 × published rate ($50/M out Fable, $25/M Opus). GPT-5.5 and the GLM route: real usage tokens at their at-cost gateway rates.

Labels withheld on purpose

We don't map model to letter. If quality were the differentiator you'd spot the expensive one. You can't — so it isn't.

Ran 2026-07-04 · Two models ran as Claude agent runs via the Anthropic API; GPT-5.5 and one more ran through our at-cost gateway. Same prompt, one shot each.

Send us your AI bill.

We route every call to the cheapest model that clears your quality bar — and prove it nightly. What you spent, what you'd have spent, receipts attached.