Liquid Inference vs Requesty

Requesty is a hosted gateway. You pick a provider and a model, or you write a routing policy, and Requesty adds 5% when you buy credit. Liquid Inference is an exchange. Providers quote their own prices, and each request starts an auction among the offers that meet your rules.

Both services take the OpenAI and Anthropic formats. The facts about Requesty come from its own pages, checked on 10 November 2026. The links are in Sources.

Feature by feature

Feature	Requesty	Liquid Inference
Price	The provider's price	Each provider quotes its own price
Provider choice	You name it, or set a fallback, split or fastest policy	An auction: the lowest charge plus wait cost wins
Fee	5% on each top-up	5% of the charge, or a floor on cheap models. No top-up fee
Cap for each request	No	Yes. `cap_usd`; the provider pays any usage above it
Spend limits	Yes. Monthly, for each key, service account and user. Group budgets need Enterprise	Yes. For each key, workspace, organization and end user
Preview with no charge	No	Yes. The charge and cap of each offer, and why others are out
Record of each request	Yes. `usage.cost` in each response, and a log	Yes. A receipt: rates, fee, cap, charge split and the offers left out
No prompt logs by default	No. Kept 30 days in the EU unless you turn them off for a key	Yes. The exchange stores no prompt or answer
No retention by default	Partly. ZDR for the organization, on written request	Yes. Only offers that keep no prompt compete
Open to sellers	No. Requesty picks the providers	Yes. By application; sellers post their own offers
Own cloud keys	Yes. No fee, for six providers	Partly. Beta. The exchange fee at the list price, for sixteen clouds
Self-host	No	No

What the exchange does that Requesty does not

Prices from competition. Each provider quotes its own price for each model, and the auction buys the lowest cost. See Price and wait.
A hard cap for each request. Send exchange.cap_usd, or the x-liquid-cap-usd header. An offer whose worst case is higher does not compete. See Purchasing rules.
A dry run. See which offers compete and what the request can cost before you buy. See Purchasing rules.
No prompt logs. The exchange keeps metadata and receipts, not the text. See Data handling.
An open market for sellers. Any provider can apply and quote. See Become a provider.

When Requesty is the better choice

You send most traffic through your own keys. Requesty charges no fee on them, at any volume.
You need EU data residency. Requesty runs its gateway in Frankfurt on every plan.

Switch in one line

 client = OpenAI(
-    base_url="https://router.requesty.ai/v1",
+    base_url="https://router.inference.ai.exchange/v1",
     api_key=...,
 )

Then change each model to a name that GET /v1/models lists. The exchange has no provider/model names. To limit the providers, use exchange.candidates.providers.allow. See Purchasing rules.

Questions

Is the fee the same?

On models that cost more than about $0.20 for each million tokens, yes: both are 5%. On cheaper models, the floor of the exchange fee can be more. Requesty adds its fee to each top-up. The exchange takes it from each settled job, so a job that fails before output pays no fee.

How do I copy a Requesty fallback policy?

Send a basket: architect.basket with exchange.candidates.models.allow. The exchange ranks the offers of each model in the list. See Purchasing rules.

Sources

Checked on 10 November 2026.

Product and routing: llms.txt, Managed policies, Latency routing
Fees and BYOK: Pricing, Billing, Bring your own keys
Spend limits: API limits, Users
Request record: Cost tracking
Logs and retention: Security, Privacy
Self-hosting and EU: Enterprise