Private alpha · Pricing

From first prediction to production, at record speeds.

Start by paying only for the rows you score. Move to the platform when you want your use cases evaluated, deployed and monitored for you.

Save 15% paying annually on Platform and Enterprise

Pay as you go

API + SDK
$0.05
per 1K rows scored

Call the API or the SDK with your history and the rows to score. You pay for the rows, nothing else.

EnvironmentsProduction
  • ✓First 100K rows every month are free
  • ✓Then $0.05 per 1K rows, $99 / month minimum
  • ✓Up to 10M rows / month
  • ✓Batch scoring on 3 tabular foundation models, on our GPUs
  • ✓Any number of use cases
  • ✓Continuous updates to the latest foundation models

No championship, safe deploys, observability or real-time endpoints.

Platform

Recommended
$1,275
per month, billed annually

Your use cases in production: evaluated, deployed safely, monitored and kept current.

EnvironmentsProdDev
  • ✓3 capabilities included, +$250 / month each extra (adds 1M rows)
  • ✓10M rows scored / month included
  • ✓Histories up to 10K rows per capability
  • ✓Weekly championship + re-run on drift
  • ✓Shadow, canary, A/B and one-click rollback
  • ✓Quality, drift and business-metric observability
  • ✓Continuous updates to the latest foundation models
Platform + forward deployed engineer$1,870

15 hours a week with an engineer who models your problems with your team and trains them on tabular foundation models. +$595 / month on top of Platform.

Real-time endpoints: ask for pricing · $0.05 per 1K rows over 10M.

Enterprise

Dedicated
from$4,250
per month, billed annually

An entire state-of-the-art ML team for the price of a single engineer: ours sit with yours, model the problems and run them in production.

EnvironmentsProdDevStaging
  • ✓Forward deployed engineers who model your problems with your team and own the results
  • ✓GPU serving tuned by foundation-model engineers: batching, multi-GPU, FlashAttention
  • ✓5 capabilities included, +$250 / month each extra (adds 1M rows)
  • ✓Fine-tune foundation models on your own use cases
  • ✓Distill foundation models into lightweight models for realtime endpoints
  • ✓Training for your team on tabular foundation models
  • ✓Dedicated capacity sized to you, with SLA, security review and SSO
  • ✓Continuous updates to the latest foundation models

Need it in your own AWS or GCP account? Deploy in your cloud (BYOC) on request.

Open source SDKApache 2.0Free forever on your own machine: no account, no API key, no rows billed.

View on GitHub

Design partners get 50% off the first 6 months, in exchange for a historical dataset, a case study and access to outcomes so we can measure lift together.

What it costs in practice

Monthly churn scoring
200K customers scored once a month
Pay as you go$99 / month
Insurer
Churn, cross-sell and claims fraud in batch, 5M rows a month
Platform$1,500 / month
Fintech
Fraud at checkout in real time, plus 2 batch use cases
Platform + 1 real time$1,500 / month
+ real-time quote
Retailer
3 use cases in batch, with a forward deployed engineer 15 hours a week
Platform + FDE$2,200 / month
Bank
5 use cases, fraud in real time, 30M rows a month
Platform + 2 caps + 1 real time + 18M rows$2,900 / month
+ real-time quote

Estimate your bill

Change the numbers to match your use cases.

Platform
$1,500
Extra capabilities
$0
Real-time endpoints
Ask for pricing
Rows over 10M
$0
Forward deployed engineer
$0
Platform, estimated
$1,500 / month
$18,000 a year, or $15,300 paid annually
Rows only, without the platform: $245 / month on pay as you go.

Why it's priced this way

Keep up with the state of the art
Tabular foundation models are improving fast. Every new one is evaluated on your data and rolled out only when it wins, so you stay current without re-platforming.
Priced on capabilities in production, not on GPU hours
Serving a model is only the start: it has to be evaluated, deployed safely, monitored and kept current. You pay for use cases that stay healthy in production, and we run everything underneath.
No training, so you only pay for inference
Tabular foundation models use your history as context instead of learning weights from it. There are no training runs, and a new use case is live in minutes, not quarters.
Three use cases included, because one is rarely enough
Each target (churn, purchase propensity, fraud) is its own capability with its own evaluation, drift monitoring and endpoint. Three come with the platform, so the second and third aren't a new purchase.
Real time is an add-on because the GPU never sleeps
Batch runs on GPUs that scale to zero. A real-time endpoint keeps your history encoded in GPU memory 24/7 so it answers in milliseconds, and that reserved capacity is what the add-on covers. It is priced to your traffic, so ask us for a quote.