Sthānika AI
// 01 — sieve for enterprises

Bring us one decision. See the numbers in ten business days

Sieve is our family of open-weight decision models. You pick one decision your teams or agents make today. We score it on your past cases and show accuracy, calibration, speed and cost side by side with what you use now.

$ book_benchmark →Visit the Sieve page →
your scorecardday 10
measureyour modelsieve
accuracy
calibration error
p95 latency
cost / 1,000 decisions
Blank until we run it, then filled in on your data.
// 02 — the offer

Three steps, each with an exit you control

  1. step 1

    Decision benchmark

    free

    One decision, up to 2,000 of your past cases, scored against your current model. Results in ten business days. A mutual NDA is all it takes.

  2. step 2

    Paid pilot

    $40K

    Six weeks, up to five decisions, success criteria agreed on day one. The fee is credited in full toward Foundry or Vault, and waived if we miss the criteria.

  3. step 3

    Founding customer terms

    first five

    For organizations that sign by 15 December 2026: the Assurance monitor free for a year, a two-year price lock and a seat on the product council.

// 03 — why sieve

Fast, low-cost decisions you can audit and own

Stop paying LLM prices for every decision. Sieve reads the input once and returns a probability for every allowed answer: no free text, nothing to parse. Every claim below traces to a published evaluation run.

71.1

Sieve-31B-v3 on the Decision Index, tied for first among 237 models and ahead of Jev 1.14 at 60.24.

results
0.5%

Of Sieve-4B's answers given at 95% confidence or more, the share that were wrong. Calibration you can route on.

results
50.5 ms

Answer time for Sieve-9B-v2, in about 20 GB of GPU memory.

results
yours

Open weights you can run in your VPC, on-premises or air-gapped. Your data stays with you, and the custom models you train are yours.

models
// 04 — what you can buy

Five offerings, one decision stack

Each stands alone. Most teams start with a Foundry model running on Vault, governed by Assurance. Full detail is on the Sieve page.

  • Sieve OpenOpen-weight decision models from 0.8B to 31B, with a Jev-compatible API.available on hugging face
  • Sieve VaultA licensed container for your VPC, on-premises or air-gapped, with monthly calibration checks.founding customers now
  • Sieve FoundryA custom model trained on your rubric and decision history. You own the adapter.taking founding customers
  • Sieve AssuranceCalibration, drift monitoring and a per-decision audit log for model risk.v1, november 2026
  • Industry Decision CanvasPrebuilt, typed decisions for your industry, financial services first.financial services, q1 2027
// 05 — how the ten days work

From NDA to scorecard

  1. days 1 to 2

    Mutual NDA. We agree the decision, the answer set and how success is measured.

  2. days 3 to 7

    We run Sieve on your past cases, in your cloud or on redacted samples, and calibrate it.

  3. days 8 to 10

    We walk you through the scorecard and a go or no-go recommendation for a pilot.

What you bring

  • One decision and the person who owns it
  • Up to 2,000 past cases with known outcomes
  • Your current model or process, as the baseline
  • A cloud account, or redacted samples
// 06 — book

Book your benchmark

Tell us the decision you want to test. The first ten organizations get the benchmark free.

Prefer to write directly? hello@sthanika.ai