What is SF·SIX?
SF·SIX (San Francisco Super Intelligence Exchange) is ColdAI’s exchange for AI models: a source-backed directory of AI model prices, benchmarks and usage, prediction markets on AI model prices and usage, and model token offerings, with trading on the Hedera testnet using test funds.
Does SF·SIX use real money?
No. Trading, model token offerings and prediction markets run on the Hedera testnet with test funds that have no real value. Real-money trading, custody and mainnet actions remain separately gated and are not offered.
Do I need a crypto wallet to use SF·SIX?
No. You sign in with your ColdAI account and SF·SIX manages the Hedera testnet side for you, so there is no wallet, seed phrase or gas token to set up.
How do model token offerings work?
A creator launches a model token for an AI model they serve, and one model token pays for one input or one output AI token of that model. New offerings launch on a bonding curve whose price rises as tokens are bought, toward a target raise before a deadline chosen by the creator, with the terms (supply, curve, deadline and creator royalty) fixed at launch and bound into a terms hash. Tokens can be bought (and, on a curve, sold back) right away, but model use opens only when the offering's open market opens: after the bonding curve graduates, or once a fixed-price sale is fully sold and settled.
Browse model token offeringsLaunch a model tokenHedera Token Service (on coldai.org)Asset tokenisation (on coldai.org)
What happens when a model token offering ends?
If the curve sells out, or the deadline passes with at least the minimum raise, the offering graduates: unsold tokens are burned, a 2.5% graduation fee is taken, a platform-set share of the remainder (20% by default) is released to the creator, and aftermarket trading opens at the last curve price. If the minimum raise is not reached, the reserve is refunded pro rata to holders and their tokens are burned.
How do prediction markets on AI models resolve?
Before you trade, each market shows its exact question, named data source, observation window, cutoff and versioned resolution rule. Per-model price markets settle on the mean standard USD price per million tokens across the providers listed by OpenRouter and the Hugging Face router on the close day, and a missing source capture voids the market. A proposed result is not final: challenges or missing and conflicting evidence can delay or void resolution.
What fees does SF·SIX charge?
Bonding-curve trades pay a 1% platform fee and a 0.3% creator fee, and a graduating offering pays a 2.5% graduation fee. Aftermarket resales carry the creator royalty fixed at launch (0–50%) and a small platform royalty as native Hedera token fees, and Hedera network fees are passed through at cost plus 20%. Inference through the SF·SIX API is priced at provider cost plus 20% provider fees plus a 10% platform fee.
What does a verification badge mean?
A verified badge marks an official AI developer or inference provider account. It is granted only after an SF·SIX operator reviews a claim from someone with a verified company-domain email, never inferred from a name, handle or logo, and it is re-checked regularly and can be revoked.
Where do SF·SIX prices and benchmarks come from?
From named public sources including the OpenRouter catalog, Hugging Face Inference Providers, fal, Artificial Analysis, models.dev and LiteLLM. Every price is a dated observation rather than a firm quote, and a missing price means unknown, not free.
What is NATDAQ?
NATDAQ (National Association of Token Dealers Automated Quotations) is SF·SIX’s family of AI model usage indexes, rebased to 100. The indexes measure observed token activity within a named source; they are not prices, investable securities or a measure of global AI demand.
Does SF·SIX verify my identity (KYC)?
Only if you choose to. Creators who launch a model token offering can optionally verify their identity (KYC, a government ID and a liveness check) or their business (KYB) through Didit’s hosted flow to earn a verified badge. Didit processes the documents; SF·SIX stores only the decision, its dates and, for a business, its name and country, never documents, images, dates of birth or document numbers.
Who runs SF·SIX?
SF·SIX is built and operated by ColdAI LLC. Its research uses the same ColdAI account, API and model integrations as Medusa, ColdAI’s model playground, and use of SF·SIX is governed by ColdAI’s terms and privacy policy.
About ColdAI
SF·SIX is a ColdAI product. Read about ColdAI, browse ColdAI’s research publications, or try any model in Medusa, ColdAI’s model playground. The ColdAI homepage covers the company’s wider work.