Architect introduces liquid inference, a dynamic pricing platform for LLM services

Architect Financial Technologies has introduced Liquid Inference, a novel platform that enables real-time bidding for large language model (LLM) inference services. The platform operates as an auction system where service providers compete to offer the lowest price that meets the buyer's criteria. Developers can integrate the service by simply changing a base URL, retaining their existing code, and allowing providers to compete on cost. Liquid Inference, developed by a trading firm rather than an AI lab, leverages Architect's experience in financial exchanges to create a two-sided market for LLM inference. The platform allows buyers to set cost caps, time limits, and other constraints, while providers can update their pricing based on their own costs. Payouts are processed through Stripe, with detailed records of each job. Harrison, from Architect, noted that new providers can be onboarded in minutes, not weeks. The platform also offers live market data, including order books and transaction records, which is rare for an LLM API. Providers use the OpenAI API standard to register models and quotes, and the system supports multi-modal capabilities through preset routing rules.