zkML cost calculator: worked examples

Proof Plan estimates proving compute and onchain verification costs from your own measurements. It checks a per-job latency target and preserves model, input and verification-key hashes. It plans an integration; it does not generate or verify a zero-knowledge proof.

Three fictional scenarios, calculated by the same engine as the tool. They illustrate behavior and failure cases, not customer outcomes or measured provider performance.

Plan 100 proofs

Inspect compute, gas, latency and artifact versions together.

Estimated total USD: 50.666667
Per-job latency (s): 12.025
Budget / latency: Within / Within

Budget components
Component	USD estimate
Proving compute	0.666667
Onchain verification gas	50
Total	50.666667

Run this scenario Input JSON Calculated report

Inspect every input

{
  "statement": "Fictional demo model executed on the declared input",
  "modelHash": "sha256:1111111111111111111111111111111111111111111111111111111111111111",
  "inputHash": "sha256:2222222222222222222222222222222222222222222222222222222222222222",
  "verificationKeyHash": "sha256:3333333333333333333333333333333333333333333333333333333333333333",
  "verifierVersion": "example-verifier-1",
  "jobs": 100,
  "proveSeconds": 12,
  "verifyMs": 25,
  "proofBytes": 2048,
  "gasPerVerify": 250000,
  "gasPriceGwei": 1,
  "nativePriceUsd": 2000,
  "computeHourlyUsd": 2,
  "budgetUsd": 60,
  "latencySlaSeconds": 15
}

Gas price increases

A different gas assumption can break an otherwise feasible budget.

Estimated total USD: 150.666667
Per-job latency (s): 12.025
Budget / latency: Over / Within

Budget components
Component	USD estimate
Proving compute	0.666667
Onchain verification gas	150
Total	150.666667

Run this scenario Input JSON Calculated report

Inspect every input

{
  "statement": "Fictional demo model executed on the declared input",
  "modelHash": "sha256:1111111111111111111111111111111111111111111111111111111111111111",
  "inputHash": "sha256:2222222222222222222222222222222222222222222222222222222222222222",
  "verificationKeyHash": "sha256:3333333333333333333333333333333333333333333333333333333333333333",
  "verifierVersion": "example-verifier-1",
  "jobs": 100,
  "proveSeconds": 12,
  "verifyMs": 25,
  "proofBytes": 2048,
  "gasPerVerify": 250000,
  "gasPriceGwei": 3,
  "nativePriceUsd": 2000,
  "computeHourlyUsd": 2,
  "budgetUsd": 60,
  "latencySlaSeconds": 15
}

Proving takes longer

Longer proving affects compute cost and the per-job latency target.

Estimated total USD: 51.666667
Per-job latency (s): 30.025
Budget / latency: Within / Over

Budget components
Component	USD estimate
Proving compute	1.666667
Onchain verification gas	50
Total	51.666667

Run this scenario Input JSON Calculated report

Inspect every input

{
  "statement": "Fictional demo model executed on the declared input",
  "modelHash": "sha256:1111111111111111111111111111111111111111111111111111111111111111",
  "inputHash": "sha256:2222222222222222222222222222222222222222222222222222222222222222",
  "verificationKeyHash": "sha256:3333333333333333333333333333333333333333333333333333333333333333",
  "verifierVersion": "example-verifier-1",
  "jobs": 100,
  "proveSeconds": 30,
  "verifyMs": 25,
  "proofBytes": 2048,
  "gasPerVerify": 250000,
  "gasPriceGwei": 1,
  "nativePriceUsd": 2000,
  "computeHourlyUsd": 2,
  "budgetUsd": 60,
  "latencySlaSeconds": 15
}

A mistake worth catching

Gas-only estimates can omit L2 data fees, failed proofs, setup and queues. The displayed total covers the entered model; compare it with actual measured bills.

Use the method with your records

Estimate proving compute and one onchain gas verification per job. Compare cost and per-job latency with supplied limits, preserving hashes and verifier version.

No ZK proof generation or cryptographic verification. File hashing verifies bytes only; it is not proof of model correctness.

Read the complete method, sources and input contract. Updated 2026-09-19; by the AGI Scorecard team.

Continue your review

Compute LensBring measured workloads and account for setup, egress, storage and failed outputs before choosing compute.Data PermitCheck intended AI use against recorded grants, expiry and regions, then model a revenue split without moving funds.