Liquid Inference vs RouterPlus

RouterPlus is a hosted gateway. You name a model, and RouterPlus sends the request to the cheapest provider of that model at its list price, with no fee. 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 RouterPlus come from its own pages, checked on 10 November 2026. The links are in Sources.

Feature by feature

Feature RouterPlus Liquid Inference
Price One list price. Self-serve providers set their own Each provider quotes its own price
Provider choice The cheapest first, by input plus output price An auction: the lowest charge plus wait cost wins
Picks the model for you Partly. A waitlist; the evaluation search is paused Yes. liquid.auto picks by published score and charge
Fee None on tokens 5% of the charge, or a floor on cheap models. No top-up fee
Cap for each request No. max_price caps input and output rates, not cache or reasoning Yes. cap_usd, cache and reasoning included; the provider pays any usage above it
Spend limits Yes. Monthly, for each key and organization Yes. For each key, workspace, organization and end user
Preview with no charge Partly. POST /v1/route lists the providers in order, with no price Yes. The charge and cap of each offer, and why others are out
No charge for a failed attempt No. You pay each attempt that the provider billed Yes. No output, no charge
No prompt logs by default No. Kept up to 30 days unless the organization opts out Yes. The exchange stores no prompt or answer
Open to sellers Yes. By application and a conformance test Yes. By application; sellers post their own offers
Own cloud keys Yes. No fee, for OpenAI, Anthropic, Azure and Bedrock Partly. Beta. The exchange fee at the list price, for sixteen clouds
Reserved capacity Yes. Across providers, on one contract, for 1 week to 12 months Partly. Beta. A monthly pool from one provider
Embeddings and audio No Yes. Embeddings, speech and transcription

What the exchange does that RouterPlus does not

When RouterPlus is the better choice

Switch in one line

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

Then change each model to a name that GET /v1/models lists. See Connect an app.

Questions

Does the exchange try another provider when one fails?

Yes, but only before the first output. After that, it never switches provider, so you never get one answer from two providers. RouterPlus has the same rule.

Does the exchange pick the model for me?

Only if you ask. Send liquid.auto as the model. It weighs the published score of each model against its estimated charge. See Auto.

Sources

Checked on 10 November 2026.