01 / model family
two models.
one financial focus.
fx-1 is our State-of-the-Art frontier model for finance and demanding mathematics. fx-1 lite brings highly capable financial and mathematical reasoning to economy pricing.
- fx-1
- 3.4Ttotal parameters
- ~128Bactive / token
- fx-1 lite
- 2.4Ttotal parameters
- ~49Bactive / token
total parameters and active parameters per token describe different aspects of the model. these figures do not establish inference speed, a context limit, or the quality of a particular answer.
read the model field guide↗
02 / plan your access
start with the work.
model subscriptions are planned usage tiers. select the model and the monthly tier that best describes your intended work, then register your interest.
define the question.describe the financial or mathematical task, the source material and the result you want to review.
choose a model.consider fx-1 for frontier depth, or fx-1 lite for economy-optimised financial and mathematical work.
register your intended usage.economy and extended include fx-1 lite. frontier economy and frontier extended include fx-1. warp+ includes both models.
exact allowances, access limits and top-up details will be confirmed. subscription prices are proposed and subject to change. API token rates are separate.
explore the access desk↗
03 / token pricing
two rates per model.
input tokens are material provided to the model. output tokens are material generated in its response. the published rates below are in USD per 1 million tokens.
usd per 1 million tokens| model | input | output |
|---|
| fx-1 | $10 | $30 |
|---|
| fx-1 lite | $5 | $15 |
|---|
estimate input and output volumes independently. the Usage calculator combines those two amounts using the published model rates; it does not assume subscription credit or included API volume.
API integration documentation, supported interfaces and availability details will accompany release. this guide describes the published model and pricing information.
estimate API usage↗
04 / evaluation record
a result has a measure.
the evaluation atlas keeps the reported benchmark, model configuration and measure together. plain scores, percentages, Elo ratings and capability statuses are shown in their supplied units.
all reported results apply to fx-1. the HLE-Full result with dipcatcher is shown as a separate configuration. no fx-1 lite benchmark results have been supplied.
evaluation configurations and methodology will accompany the technical release.
open the evaluation atlas↗
05 / dipcatcher
the next layer.
dipcatcher is our State-of-the-Art harness for frontier autonomous trading. it is under development and will release soon; join the waitlist for now.
when available, dipcatcher will be bundled free with model subscriptions until the research preview ends on december 12, 2026. model subscriptions and API token usage retain their own pricing.
join dipcatcher waitlist↗
06 / terminology
the small print,
made readable.
total vs active parameters
total parameters describes the full model parameter count. active parameters per token describes the approximate parameter count used for a token. fx-1 has 3.4T total and ~128B active; fx-1 lite has 2.4T total and ~49B active.
per million tokens
1M tokens is the pricing unit used for API input and output. input and output are charged at their respective model rates. the website does not specify a conversion between tokens and words.
subscriptions vs API rates
monthly subscriptions describe planned model usage tiers. API token prices describe input and output usage separately. exact subscription allowances and included usage are still to be confirmed.
research preview
the current calls to action register waitlist interest. dipcatcher is still under development and, when available, will be bundled free with model subscriptions until december 12, 2026.
reported evaluations
the site shows company-reported fx-1 results in their original measures. a reported score is one item in the evaluation record; evaluation configurations and methodology will accompany the technical release.