Pricing
Who will check this number after you?
A validator, a governance council or a customer's reviewer will ask how sure you are. A weak answer can cost a finding, a delayed release or the wrong model in production. Pick the help you need. We set the fee with you on a 20-minute scoping call.
Three ways to start.
You choose the precision: how small a difference between two versions of your model the test can reliably detect, on our measurement scale. A finer setting needs a larger calibrated question bank, more calibration runs and more checking, so it costs more and takes longer. Each fee follows the precision, the turnaround and the scope you pick.
Start here
Independent challenge
Fee set on a 20-minute scoping call
One evaluation, one decision, your data
- For second-line model-risk and validation teams, clinical AI governance councils, and AI vendors preparing for a health system's review.
- We check the judge behind one evaluation you already run, on your data or inside your environment.
- You get a findings note with a one-page summary for your committee, and a debrief.
- Before work starts, you choose the turnaround and how small a difference between two versions of your model the test must reliably detect.
- Not a certification.
Measurement program
Fee set on a 20-minute scoping call
Your models and scales, measured again when you change a model or a judge
- For teams that need the same number to hold across model and judge changes.
- Your LLM judges and your human reviewers on one ruler, each reading dated, with its error bar.
- Hosted by us or run inside your environment, both early access by request.
- The fee follows what you choose: how small a difference the test must reliably detect, the turnaround, the number of models and scales, where it runs, and how often we measure again.
Learn
Free live lesson
Thu Jan 7, 2027, 9:00 AM ET, on Maven
- For eval builders, and for anyone on your team.
- Advanced AI Evaluation: Beyond Pass Rates.
- A free reading guide to start with before the lesson.
- Paid workshops open on Maven once their dates are set.
Hosted AIM access opens by request while judges for levels 7 to 13 are still being credentialed. Join the access list.
How buying works.
- Scope. A 20-minute call, or an email to matt@xlnc.co. We confirm the evaluation, the decision it supports, the precision you need and who will review the result.
- Agreement. An NDA before any non-public material, then a statement of work with the fee, the precision, the turnaround and the deliverable.
- Access. You supply the evaluation data, or access to run inside your environment with the model endpoints you pick. You pay for your model's calls.
- Invoice. Fees are set in your order. Unless your order says otherwise, they are due within 30 days of invoice and exclude taxes.
- Delivery. The findings note, the record behind each number, and a debrief with your team.
Security and data.
Your security review gets written answers, item by item. From us: the published method, the engine source on request under a reviewer agreement, and a plain inventory of every data input, where it lives and which providers process it. From you: your questionnaire and your data rules. We will not quote a timeline we cannot keep.
If your data must stay with you, the harness is designed to run in your own environment, with the model endpoints you pick, so your traces do not pass through our servers. Early access, by request.
We do not sell what you send. Our Privacy Policy names every provider that processes it.
Good fit, and not a fit.
- Good fit
- A model or AI evaluation is about to face a validator, a governance council or a customer's review. You can share the evaluation data, or let us run inside your environment. You want a written record, not a verbal opinion.
- Not a fit
- You need a certification, an accreditation or a regulatory opinion: we give none. Your model changes daily and cannot be held to one version for the check. You want to learn the method yourself: start with the free lesson.
Questions about price.
- Why are there no monthly plans?
- Hosted AIM opens by request while judges for levels 7 to 13 are still being credentialed. Until then we price the work you need, not a quota of runs.
- What sets the fee?
- The precision you choose (how small a difference between two versions of your model the test can reliably detect), the turnaround, how many models and scales, where the work runs, and how often we measure again. A finer setting needs a larger calibrated question bank, more calibration runs and more checking, so it costs more and takes longer.
- Who pays when an AI vendor commissions the check?
- The vendor can. The statement of work names who commissioned it, and the findings note says so.
- Is this a certification?
- No. We report what we measured, with its error bar and its date. We name no accreditation we do not hold.
- What do you need from us?
- The evaluation you run today, the decision it supports, and who will review the result. For a run inside your environment: access and the model endpoints you pick.
Built on published measurement science.
The method has been published in peer-reviewed venues since 2016, including a 2024 De Gruyter chapter in a volume co-edited by Leslie Pendrill, a former Chair of EURAMET. Read the open-access chapter.
Every score has metrological traceability: it leads back to one locked question, with its answer key and its grading rule.
What number must you defend?
You will get a short note from Matt when a measurement slot opens. Nothing else, no drip sequence.
We store what you send with the few providers our Privacy Policy names, and use it to reply to you. We do not sell it. Email matt@xlnc.co and we delete it. Read our Privacy Policy.