Mercatus AI pulled all 17 third-party hosts serving the same V4 weights three weeks later and ran the workload math, finding sixteen of seventeen hosts now undercut DeepSeek's own API.

Mercatus is the first exchange for AI inference tokens. AI companies increasingly budget around one metric: the cost of inference, measured in dollars per million output tokens. Yet there has never been a market where that price could be traded directly. Mercatus changes that. Model providers and inference platforms sell forward token capacity. Enterprises, traders, and investors buy it to hedge future inference costs or express market views. Orders match continuously in a 24/7 central limit order book, with physical settlement of LLM tokens and no intermediaries. Mercatus doesn't publish prices—it discovers them. Every physically settled forward establishes a market-clearing price for AI inference. That price becomes the benchmark for pricing, hedging, and financing compute across the AI economy. The AI economy runs on tokens. Mercatus is where those tokens are priced.