Forward inference, answered.

What exactly am I buying?

A contract: a fixed 1,000,000,000-Token allocation (1 Token = 1 output token) on one model for one delivery month, at a price pinned when you buy. It connects to an API key you generate, and you call the model through an OpenAI-compatible endpoint.

Why is it cheaper than on-demand?

Commitment. The standard curve discounts 10% per 30 days to maturity (the delivery month’s last day), capped at 20% below the model’s list rate: providers sell forward against demand they can see; buyers commit early and pay less.

What happens if I don’t use the Tokens?

Unused Tokens expire when the delivery window closes. On series (a model’s listed contract months) where the provider consents to resale, an untouched contract can be relisted on the order book before its first consumed token. Buy what you’ll use.

Do I have to change my code?

No. The API is OpenAI-compatible: point your client’s base URL at Touchmark’s router, use your key, keep your code.

How do input tokens and cache reads count?

1 Token = 1 output token. Input tokens meter at the model’s pinned input:output price ratio and cache reads at cache:output; at a ratio of ×0.32, one input token draws 0.32 Tokens. One allocation covers mixed traffic, and each listing states its ratios.

Are there rate limits?

Each listing states its rate caps (requests and tokens per minute) and any usage terms; they are part of the contract you see before you buy.

What are best-in-class contracts?

Benchmark-referenced contracts that name no model at sale. They are quoted in Synthetic Tokens (1 SYN = 1 Claude Opus 5-equivalent output token: equivalent in work, sized on a fixed benchmark set) and resolve about 20 business days before the delivery window into the top-ranked generally-available model on the series’ named public index (the Artificial Analysis Coding Index, for the best-coding series) under a rule pinned at listing. The allocation then converts into that model’s Tokens by the published Conversion Factor, clamped to 0.25–4.0; if measurement fails, it settles 1:1 into Claude Opus 5 Tokens through the same listed provider.

Can I buy more than the posted book shows?

Yes: the RFQ (request-for-quote) desk. Post the size, months, and models you need; providers respond with quotes, and awarding one is binding on the spot. RFQ deals settle directly between you and the provider, off the posted book.

How do payments work?

Three ways. Fund a USD balance by bank transfer (the instructions are in the app; business verification comes first) and purchases on the book settle from it the moment you click. Or pay the seller directly when you buy: by card for a single contract, up to the card limit shown in the app, or by bank wire for any size, with the contract delivered once the wire lands. Sales are final at execution.