TLDR Tech / Crypto
Ambient: Proof-of-Useful-Work AI Inference Network
Ambient is a proof-of-useful-work inference network where GPU operators compete via reverse auction to serve two curated open-weight models: Kimi K2.7 Code (262K context, $0.75/$3.50 per 1M tokens) and GLM 5.1 (203K context, $0.98/$3.08), with rewards tied to lowest cost and latency. The chain handles job creation, auctions, bid commitments, settlement, reputation, and penalties, while a Proof of Logits method fingerprints model execution through internal activation states so validators can confirm work without re-running inference. Ambient has already surpassed Moonshot, Kimi's creator, in Kimi inference volume, and Delphi Ventures has backed the project at both pre-seed and seed. Available via Desktop app, OpenAI-compatible API, and OpenRouter, Ambient targets the role of a market-priced, censorship-resistant global serving layer for open-weight models.
Read the full story