Liquid Inference

Liquid Inference, by the AI Exchange and Architect Financial Technologies, Inc., is an auction-style marketplace for LLM inference. Users send ordinary chat requests through an OpenAI- or Anthropic-compatible API, and providers compete to answer each one at the lowest price meeting the user's rules.

What It Is

Liquid Inference is an inference routing exchange. Providers set a per-token price for each model they serve and can change it at any time. Customers choose the model, region, speed and minimum quality, and the cheapest provider that meets those rules answers the request. The site lists model families such as Qwen, GLM, GPT, Gemma, DeepSeek, Gemini, Claude, Grok, Mistral, Kimi and Llama.

How requests are priced and routed

The maximum cost of a request is fixed before the model starts, and billing covers only the tokens produced. Billing is itemized line by line. Public price history for live and past prices can be downloaded, so work can be sent when prices are low. The vendor says Architect Liquid Inference regularly tests every provider with standard benchmarks and publishes the results.

Compatibility and setup

OpenAI Chat Completions and Anthropic Messages are served on one base URL, so existing code works by changing the base URL and API key. The site lists compatible clients and SDKs including Claude Code, Codex, Cursor, OpenCode, Cline, Roo Code, Kilo Code, Goose, Open WebUI, n8n, the OpenAI and Anthropic SDKs, the Vercel AI SDK, LiteLLM, LangChain and curl.

Selling inference

Providers can sell inference on models they already run. Their software posts prices and adjusts them as costs change. The provider flow is: create an account, register a deployment, have an operator review and list it, then connect an agent and start quoting. Providers can stop quoting when full or lower prices when they have spare capacity.