Networks · 8 October 2026 · 3 min read

Architect starts auctioning each AI prompt

Liquid Inference takes a request, collects quotes, and sends it to the cheapest provider that meets the buyer’s rules. AkashML is a launch partner. The usual SDKs still work.

Launch graphic for Architect’s Liquid Inference, listing AkashML among the partners. — Illustration. Not a photograph of the machines or the event in the story.

Architect has launched Liquid Inference, @MechEconomy reported. Each prompt is auctioned. The lowest quote that meets the buyer’s rules wins. OpenAI and Anthropic’s own software kits still work, behind one API key, with a live price instead of a printed rate card. AkashML is a launch partner. The launch graphic also names Boundless, Engy, Grizzly, Hyperbolic, Pearl Research, and Zro.

This is the version of decentralized compute that would matter to a robot or an app: the spare GPUs sit inside the request, not on a separate marketplace the developer has to learn. Whether the winning price is actually lower, and whether the answer still arrives in time, is not in the launch note. A router is not a proof of savings.