Enter the market
Backed by Y Combinator

touchmαrk

Enter the market
The problem

AI capacity has no market. Every reserved deal is a private negotiation.

Today a reserved-capacity deal is negotiated privately: the price depends on who's asking, and the commitment can't be resold. Power, oil and grain trade forward on open markets. AI capacity doesn't yet.

For companies

Once you commit, you're stuck.

Committing forward is how you get a real discount, so companies commit. But there's no exit: if plans change or usage drops, the contract stays and you keep paying for capacity you won't use. You want the discount and a way out; today you can't have both.

For providers

GPUs are financed on a guess.

A GPU buildout is committed years ahead; the revenue that pays for it arrives month by month. Providers can't see forward demand and face limitations in future capacity sales, so there's no revenue locked against the spend, and nothing to plan the next buildout on.

The fix

Give it a market.

Touchmark puts reserved capacity on a live order book. Buyers get the discount at a posted price; providers sell forward against demand they can see.

The market

Discover the price of intelligence.

Standard capacity contracts on a live order book. The open market gives both sides a better deal than a private negotiation.

For buyers

Commit early. Pay less.

Buy inference capacity in advance at a significant discount: a fixed block of tokens on a named model for a delivery month, served through one OpenAI-compatible key. Draw it down at your pinned per-token rates from the month’s first day.

  • Up to 20% off list for committing ahead
  • A posted price, pinned the moment you buy
For providers

Sell forward. Fund the buildout.

Sell capacity in advance to lock in revenue up front and maximize GPU utilization; idle hours can find buyers before the month starts. The order book shows demand at each delivery month, so GPU commitments are planned against orders already in hand. You set the price and the floor; delivery stays on your endpoints.

  • Revenue committed at sale, not at usage
  • Demand visible by delivery month
  • Settlement handled by the market
The instruments

Contracts for both sides of the market.

Volume contracts and the floating Top-1 series are live today; named pre-release series are coming soon.

GLM52·1B·SEP26#07 LIVE BOOK
11circ
Remaining
1,000M
OF 1 B TOTAL

GLM 5.2

Model
GLM 5.2
Volume
1 B
Window
Sep 1 – Oct 1
Scope
Tokens
Delivery
Connects to your API key
usable from Sep 1
Settlement
BALANCE
Price
$3,989.33
What you get
A fixed 1,000,000,000-Token allocation on GLM 5.2: 1 Token = 1 output token; input counts at ×0.32, cache reads at ×0.06. You connect it to your API key and call the model as usual.
Pricing
Bought at the live best ask on a two-sided book. September quotes $3,989.33 against $4,400 at list; further months fill deeper down the curve.
Settlement
Funded from your USD balance: a synchronous purchase, the price pinned the moment you click. Sales final.
Resale
Live on the market for consenting providers; pending on this shelf. Buy what you’ll use: unused Tokens expire when the delivery window closes.
KIMIX·SU·DEC26#12COMING SOON
0SUcirc
Remaining
10,000SU
OF 10,000SU TOTAL

Kimi-X

Model
Kimi-X · unreleased
Volume
10,000SU
Window
Dec 1 – Jan 1
Scope
synthetic units
Quoted in
Synthetic units of work
settles into Tokens at launch
Settlement
INTO TOKENS
Price
$0.29 / unit
Quoted in
Synthetic units: a cross-model unit of work, sized on a fixed, versioned benchmark set. You can’t quote tokens on a model nobody has run; you can quote the work.
Settles
At launch, Kimi-X runs the same set, its token appetite is measured, and each unit converts into a concrete Token allocation on the model.
Either way
A hungry model settles into more Tokens, an efficient one into fewer; the work you hold is constant.
Sizing
Your Tokens per month ÷ your current model’s per-unit appetite = the units you need. 10,000 units ≈ a 1B-Token month on Kimi K3 Fast, today’s yardstick.
K3F·1B·NOV26#03 LIVE BOOK
4circ
Remaining
1,000M
OF 1 B TOTAL

Kimi K3 Fast

Model
Kimi K3 Fast
Volume
1 B
Window
Nov 1 – Dec 1
Scope
Tokens
Delivery
Connects to your API key
usable from Nov 1
Settlement
BALANCE
Price
$18,000
What you get
A fixed 1,000,000,000-Token allocation on Kimi K3 Fast, the premium block on the shelf: 1 Token = 1 output token; input counts at ×0.20, cache reads at ×0.02.
Pricing
The far end of the listed curve: $18,000 for November against $22,500 at list — the 20% early-commitment discount, the curve's cap.
Settlement
Funded from your USD balance: synchronous, the price pinned at click. Sales final.
Resale
Live on the market for consenting providers; pending on this shelf. Buy what you’ll use: unused Tokens expire when the delivery window closes.
TOP1·CODE·1B·OCT26#21 LIVE BOOK
0SUcirc
Remaining
1,000MSU
OF 1 BSU TOTAL

Top-1 Coding Model

FLOAT
Model
Floating — named before the window
Volume
1 BSU
Window
Oct 1 – Nov 1
Scope
synthetic units
Delivery
Whichever model leads the coding index
named before the window opens
Settlement
INTO TOKENS
Price
$20,083.33
What you get
A fixed 1,000,000,000-synthetic-unit allocation on the market’s best coding model: you commit to #1 today; which model that is gets decided later.
Selection
A public rule pinned at listing: the top of the published coding index is named about 20 business days before the delivery window opens, and your allocation converts into that model’s Tokens by its measured efficiency.
Pricing
$20,083.33 against $25,000 at list: the early-commitment discount, deepest far from maturity. 1 synthetic unit = 1 Claude Opus 5-equivalent output token.
Resale
Allowed on this series: an untouched contract can be relisted on the book any time before its first consumed token.
01 / 04GLM52·1B·SEP26
Illustrative cards. Terms and prices are not live data.
The quote

The price is a curve.

The market below is illustrative: 1B-token blocks on one model, one delivery month each. Committing ahead earns the discount. The further the delivery month, the deeper the price below list, up to the curve's 20% cap; the discount shrinks as a month nears maturity.

Forward curve · best ask

ⓘ

Best ask by delivery month: SEP26 $3,989.33, OCT26 $3,534.67, NOV26 $3,520.00.

Delivery month

Starts 1 Sep 2026 · ends 1 Oct 2026

Order book · Sep 1 – Oct 1

ⓘ
PriceUnitsTotal
Asks
  • $4,079.006$24,474.00
  • $4,039.004$16,156.00
  • $4,014.001$4,014.00
  • $4,001.003$12,003.00
  • $3,989.002$7,978.00
Last $3,989.00Spread $52.00
  • $3,937.002$7,874.00
  • $3,924.003$11,772.00
  • $3,905.005$19,525.00
  • $3,879.004$15,516.00
  • $3,847.008$30,776.00
Bids

Example: what a 1B-token block yields on your traffic. Drag the handles to your expected raw-token mix.

1 Token = 1 output · input ×0.32 · cache read ×0.06
Expected traffic mix

Input 35% ≈ 1.44B · Output 10% ≈ 410.1M · Cached 55% ≈ 2.26B — of 1.00B Tokens

SEP26 contract $3,989.33 ÷ 4.10B raw tokens ≈ $0.97 per 1M raw, at your mix.

Request for quote

Name your terms. Providers bid.

Need something the board doesn't list (a bigger block, dedicated GPUs, reserved throughput, latency floors)? Post a request with your exact terms. Providers quote against each other for it: you see every quote as it lands, they see only the level to beat. Accept the best one and it becomes a contract on the spot.

Illustrative tape. Quotes are not live data.
Before it ships

Buy capacity on models that don't exist yet.

Best-in-class contracts track the top of the arena board and resolve to whichever model is #1: you commit today, the winner is named before your delivery window opens.

Illustrative board. Scores and timing are not live data.
How it works

Best price for committing early. Pinned the moment you buy.

01Fund a balanceDeposit dollars: the wallet every purchase settles from.
02Buy the contractPick a model and a delivery month; buy at the live best ask. The price is pinned the moment you click.
03Get your keyGenerate an API key and connect your contract to it. From the month’s first day it’s live. Use it like any other key.
04Draw it downUsage draws the block down at your pinned rate through the delivery month. Buy what you’ll use; unused Tokens expire when the window closes.
Enter the marketAlready in the market? Sign in
The index

AI is a commodity. This is its index.

Every filled order is a price someone actually paid for model capacity. That makes the index publishable: one reference series per model and delivery month, normalized to spot, with synthetic-unit series for models that haven't shipped.

TOKENS
GLM52 · SEP26
90.7
▲ +0.3
TOKENS
GLM52 · OCT26
80.3
▲ +0.2
TOKENS
GLM52 · NOV26
80.0
▼ −0.2
TOKENS
GLM52F · SEP26
90.7
▲ +0.1
TOKENS
GLM52F · OCT26
80.3
▲ +0.4
TOKENS
GLM52F · NOV26
80.0
▼ −0.2
TOKENS
K3F · SEP26
90.7
▲ +0.4
TOKENS
K3F · OCT26
80.3
▲ +0.1
TOKENS
K3F · NOV26
80.0
▼ −0.1
SYNTH
KIMIX · DEC26
91.2
▲ +0.8
SYNTH
GLM6 · JAN27
88.6
▲ +0.5
Spot-normalized · spot = 100 · illustrative
Where this goes

Our vision: solve AI economy.

We want to build all the infrastructure the AI economy needs. Today that's a market for capacity. The rest is still unsolved: how AI work is priced, measured, verified, and settled. We'll take these one market at a time.

Team

Press