Liquid Inference vs LiteLLM
LiteLLM is an open-source proxy and Python library that you run yourself. It sends each request to a provider account that you configure, and you pay each provider yourself. Liquid Inference is an exchange. Providers quote their own prices, and each request starts an auction among the offers that meet your rules. You pay every provider from one balance.
The two can also work together: LiteLLM can send requests to the exchange as one more provider. The facts about LiteLLM come from its own pages, checked on 10 November 2026. The links are in Sources.
Feature by feature
| Feature | LiteLLM | Liquid Inference |
|---|---|---|
| What you buy | Nothing: the proxy is free. Enterprise is a yearly licence | Inference. One key and one balance for each provider |
| Who runs it | You run, patch and scale the proxy | The exchange. There is nothing to run |
| Price | Your contract with each provider | Each provider quotes its own price |
| Provider choice | A strategy: weighted, latency, usage or lowest list cost | An auction: the lowest charge plus wait cost wins |
| Fee | None on tokens | 5% of the charge, or a floor on cheap models |
| Cap for each request | No. Budgets for each key, user or team, off by default | Yes. cap_usd; the provider pays any usage above it |
| Preview with no charge | Partly. An estimate from token counts that you type | Yes. The charge and cap of each offer, and why others are out |
| Record of each request | Partly. A spend log, priced from LiteLLM's own list | Yes. A receipt at the rates of the offer |
| Open to sellers | No. Only the providers that you configure | Yes. By application; sellers post their own offers |
| Quality checks | Partly. A health check sends a test request to each deployment | Yes. Measured speed, and a check of each tool call and structured answer |
What the exchange does that LiteLLM does not
- Nothing to run. No server, database or upgrade.
- One balance for every provider. You need no account, contract or key at each provider.
- 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 thex-liquid-cap-usdheader. See Purchasing rules. - An exact charge. The receipt shows what you paid, from the metered usage and the rates of the offer. See Receipts.
When LiteLLM is the better choice
- You have contracts with your providers, and no data may leave your own network.
- You need a provider or a model that the exchange does not list. LiteLLM supports more than 140 providers.
Use the two together
Add the exchange to your LiteLLM config as an OpenAI-compatible provider. LiteLLM keeps its routing and logs, and the exchange runs the auction for the requests that you send to it. See LiteLLM.
To move all traffic to the exchange, see Move from LiteLLM.
Switch in one line
client = OpenAI(
- base_url="http://localhost:4000",
+ base_url="https://router.inference.ai.exchange/v1",
api_key=...,
)
Then change each model to a name that GET /v1/models lists.
Questions
Do I need my own provider keys?
No. The exchange pays each provider from your balance. You can also connect your own cloud key, in beta. See Bring your own key (BYOK).
Were LiteLLM releases compromised in 2026?
LiteLLM reported two incidents. On 24 March 2026, PyPI versions 1.82.7 and 1.82.8 contained a credential stealer. On 21 October 2026, version 1.107.0 contained malware on PyPI and on Docker. LiteLLM's own posts have the details and the fixed versions.
Sources
Checked on 10 November 2026.
- Product and pricing: Pricing, LiteLLM
- Routing: Routing
- Budgets: Users and budgets
- Cost record and estimate: Cost tracking, Pricing calculator
- Health checks: Health
- Security incidents: March 2026 update, 21 October 2026 incident