Pricing
You pay for the decision. Tokens go through at cost.
A Proving Ground run is priced per run, from what it actually cost to measure. Production traffic on platform accounts converts vendor list price into credits with no per-token markup — and traffic on your own vendor keys costs nothing at all here.
| What | How it is priced | Figure |
|---|---|---|
| Proving Ground run | Per run — the measured inference and judging cost of that run, plus a platform premium | To be determined |
| Membership | Includes routing: run your profiles from your own application through the API | To be determined |
| Credits (platform accounts) | Vendor list price converted to credits. No per-token markup — what the platform sells is the decision | To be determined |
| BYOK traffic | Your vaulted vendor keys, your vendor account. The platform takes nothing | $0.00 |
| IaaS / rented capacity | Provisioned per engagement and billed by the hour, so per-token credits do not apply until an instance is live | To be determined |
| Private Cloud (owned + certified partner) | Per-token, on named single-operator hardware | To be determined |
Why the figures are blank
What a run actually costs to produce
Every completed run reports its own economics back to you: how many answers came back, what judging cost, what the run was priced at, and how long it took. Nothing is estimated after the fact.
Answers
43 of 45 ok
Judging cost
$0.00412
Run price
$1.68
cost +18%
Duration
6 min
A real run summary. The premium percentage is an operator setting, shown to you rather than folded into the total.
Credits, plainly
Credits are the customer-facing unit for production traffic on platform accounts. The catalog shows both figures side by side — the credit price and the underlying vendor cost it derives from — so the markup, when there is one, is visible rather than implied.
| Model | Credits / 1M in | Vendor cost |
|---|---|---|
| Claude Haiku 4.5 | 80.0 cr | $0.80 |
| Claude Sonnet 5 | 300.0 cr | $3.00 |
| Gemini 3.6 Flash | 15.0 cr | $0.15 |
| Gemini 3.1 Pro | 125.0 cr | $1.25 |
Bake-offs and drift detection always rank on vendor list price, never on the marked-up figure.
Get a priced answer for your workload
Three representative items and the number you run per month is enough to produce a ranking costed at your real volume.