Our background is trading systems and exchanges. Liquid Inference applies that to LLM inference. On Liquid Inference, model providers compete with each other on an open marketplace to offer the best price for any given model. We paired that with an intelligent auto-router so you always pay the lowest cost on the spot for a preset quality level.

Currently we're cheaper on Kimi/GPT/GLM but working hard to be the best price for all models.

You can also customize your minimum routing settings if you have particular requirements like what region the compute can be served from, zero data retention, provider/model allowlists, etc.

It's easy to set up, I usually just ask the harness itself to "write me a shell script to start Claude Code using Liquid Inference" for example and it just goes from there.

We're eager to hear your feedback!