–
per share
Example
Pricing a hypothetical contract.
Price an option and check the quote against simulation.
Set the contract in the panel, or type a ticker to use a real one. Everything reprices as you change it, and a fresh simulation of the same model checks the network's price on every change.
The Model price card shows the pricer's output: one forward pass through a five-network ensemble, with the five Greeks taken as exact derivatives of that pass by automatic differentiation. The Monte Carlo price beside it re-simulates the same contract on a new seed, with a 95% interval, using the engine that produced the network's training labels. Above 12 trading days that is the control-variate Asian engine; at or below, it is the rough-volatility engine the short-dated network was distilled from. Cross-check, beneath it, is the gap between the two prices in basis points of the strike. It measures how closely the network reproduces its reference engine. It does not test the model's assumptions about the market.
The contract is an above 12 trading days. At or below 12 trading days a second model trained on rough volatility takes over and prices a standard European option. The price jumps across that boundary because the two sides are different contracts under different dynamics.
No price on this page is a market price, and none is quotable. The short-dated model's rough-volatility parameters and the Hedge tab's simulated market are fitted to SPY option quotes captured on 20 August 2026. Every other input is hypothetical unless you load a ticker in the contract panel. The Hedge tab sells a 30-day at-the-money call and hedges it daily under transaction costs, then reports the tail loss of a learned policy and of two textbook rules on the same paths.
Start from
–
–
–
Cross-check
–
–
Risk
Δ
Deltaper $1 of spot–
Γ
Gammadelta change per $1 of spot–
ν
Vegaper volatility point–
Θ
Thetaper trading day–
ρ
Rhoper rate point–
Check this quote
Both panels recompute as you change the contract.
Price against simulated paths
The simulation's estimate and its 95% band as the path count grows, with the model price as the dashed line. Use it to decide how many paths this quote needs; the count is set in the contract panel.
Method
These runs are seeded, so the curve is reproducible; the high-precision reference uses an independent seed.
Price attribution
How much each input adds to the price, starting from a minimal at-the-money option.
Method and checks
Integrated Gradients against an at-the-money baseline option. The contributions plus the baseline price reproduce the quoted price; the residual is the discretisation error of the integral.
Measured accuracy of the model price
Pricing accuracy for the averaged contract, out of sample
Measured against high-precision simulation on averaged contracts held out of training and drawn from the same parameter box, so this is the model's error inside that box. The short-dated model is measured separately, on the methodology page.
Response time on this server
Wall-clock time for each pricing task on this shared instance, measured when you press Time it. Timings on a shared instance vary from run to run.
Browse the averaged contract's range Price over strike across moneyness 0.55 to 1.95 and expiries from 0.06 to 2 years, at the current volatility and rate. The short-dated model's range is the surface below. Click a point to price it.
Strike and expiry grid
Price divided by strike, over moneyness and time to expiry, at the current volatility and rate.
Throughput
Short-dated volatility surface
A model smile for expiries up to 12 trading days, from a surface trained with butterfly and calendar-spread penalties. Both conditions are re-checked by automatic differentiation at every point shown, and the result is reported below for this grid. The surface is a constrained fit to the short-dated pricing model and is no more accurate than that model. It depends on volatility and rate only, so it does not move with the contract in the contract panel.