Touchmark

Touchmark is a market for AI model inference where buyers purchase reserved capacity ahead of time as standardized contracts and providers bid to supply it. A contract is a fixed allocation of one billion Tokens on one model for one delivery month, where one Token equals one output token, input tokens count at the model's input to output price ratio and cache reads count at the cache ratio. Listings seen on the site include GLM, Kimi and Gemma models from providers such as OpenRelay and Wafer. It is aimed at teams with predictable inference spend who want a lower price and more certainty than buying at retail. The buyer picks a model and delivery month, buys at the live best ask with the price pinned at purchase, generates an API key and draws the contract down through an OpenAI compatible API. Unused Tokens expire when the window closes, and where a provider consents, an untouched contract can be relisted on the order book before its first token is used. The site says there is no buyer fee and describes provider pool failover. Pricing follows a published curve. Each model has a list rate, and the further ahead the delivery month is bought, the deeper the discount, capped at 20 percent below list, reached about 60 days before the month ends. Buyers can deposit dollars to a balance or pay the seller by card or bank wire. The site says Touchmark is backed by Y Combinator.