Pay for eval runs and seats, not for tokens you cannot forecast. A HIPAA BAA comes with every plan.
14-day free production trial with 5,000 runs. No card.
For seed-stage teams putting their first regulated agent under test.
per month, billed monthly
Start free trialFor Series A to C teams running several agents through CI every day.
per month, billed monthly
Start free trialFor health systems, banks and insurers with model-risk and audit obligations.
annual contract, typically $35k to $120k per year
Book a technical demo32 rows across usage, evaluation, CI gates, compliance and support. Enterprise adds deployment control and expert-curated datasets on top of Scale.
| Feature | Developer $499/mo | Scale $1,899/mo | Enterprise Annual contract |
|---|---|---|---|
| Usage and seats | |||
| Eval runs included per monthOne run is one case scored by the full judge ensemble | 50,000 | 250,000 | Custom volume |
| Overage per 1,000 runs | $6 | $6 | Contracted rate |
| Vertical benchmark packs | 2 packs | All packs | All packs + custom |
| Additional pack | $250 / month | Included | Included |
| Seats included | 3 | 10 | Unlimited |
| Additional seat | $49 / month | $49 / month | Included |
| Parallel sandboxed runnersUpper bound for max_concurrency per suite | 10 | 50 | Custom |
| Run history retention | 30 days | 1 year | Your retention policy |
| Evaluation engine | |||
| Bring your own cases (JSONL, CSV, traces)De-identified on ingest | |||
| Trajectory capture and tool-call schema assertions | |||
| Sequence, boundary and loop constraints | |||
| Calibrated judges with published kappa | Your 2 packs | ||
| Multi-judge consensus ensembles | |||
| Synthetic edge-case injector | |||
| Perturbation and prompt-injection suites | |||
| Trajectory diff timeline across prompt versions | |||
| Custom golden datasets curated with domain experts | |||
| CI/CD gates | |||
| GitHub Actions, GitLab CI and CircleCI | |||
| PR comment bot with regression diffs | |||
| Fast-path gate (20 critical cases per push) | |||
| Sharded full-suite runs on merge or nightly | |||
| Cost and latency drift tracking | |||
| Compliance and security | |||
| HIPAA BAA | |||
| SOC 2 Type II report under NDA | |||
| Compliance changelog exports (PDF, JSON) | |||
| Annex IV, GMLP and SR 11-7 report templates | |||
| SHA-256 signed run manifests | |||
| SAML SSO (Okta, Microsoft Entra ID) | |||
| SCIM provisioning and fine-grained RBAC | |||
| Deployment | US or India cloud | US or India cloud | VPC or air-gapped |
| Support | |||
| Support channel | Email, 1 business day | Shared Slack, 4 hours | Named solutions architect |
| Uptime SLA | 99.9% | ||
The scripts are cheap to start. The cost shows up every week after: fixtures drift, rubrics get rewritten by whoever touched them last, and a single-prompt judge raises false alarms that an engineer has to read case by case. In our calibration study an uncalibrated single-prompt judge reached a Cohen’s kappa of 0.42 against expert consensus, against 0.87 for a calibrated ensemble.
Kappa figures: GONATURELLE calibration study, 2026 (6,400 double-annotated cases). Monthly totals: $15,000 vs $4,899.
This counts engineering time only. It leaves out what a regression costs once it reaches patients, borrowers or claimants: remediation, notices, and exposure such as EU AI Act fines of up to €15M or 3% of worldwide turnover for high-risk obligations.
Runs keep going; nothing is throttled mid-gate. Usage above your plan is billed at $6 per 1,000 runs on Developer and Scale, metered in blocks of 1,000 and itemised on the next invoice. You get an email at 80% and 100% of the included volume, and workspace admins can set a hard monthly cap that pauses full-suite runs while keeping fast-path gates alive.
Custom curation is scoped inside an Enterprise contract, not sold by the case. Scope depends on case count, workflow depth (single turn or multi-step trajectories) and reviewer credentials (MD, JD, CPA, actuary or former examiner). Every custom pack is double-annotated, adjudicated and delivered with its inter-annotator agreement, and you own the cases.
Dedicated single-tenant VPC (AWS, GCP, Azure) and self-hosted air-gapped Kubernetes are Enterprise options on an annual contract, typically $35k to $120k per year depending on run volume, deployment model, custom datasets and support level. Your cloud compute and any model inference inside your account are billed by your provider, not by GONATURELLE.
Developer runs up to 10 parallel sandboxed runners per suite and Scale up to 50; Enterprise limits are set by your contract or by your own cluster on self-hosted deployments. The SDK's max_concurrency argument is capped at your plan limit. Fast-path gates are scheduled ahead of nightly full suites so a large backfill never delays a PR check.
14 days of the Scale feature set with 5,000 eval runs, every vertical pack and the GitHub Action. No card is required. At the end of the trial your suites, baselines and run history stay in the workspace; pick a plan to keep gates running.
Anyone who signs in to the workspace: engineers, clinical or compliance reviewers and admins. CI service accounts and PR bot identities are free. Developer includes 3 seats and Scale 10; extra seats are $49 per month. Read-only links to signed reports can be shared with auditors without a seat.
Annual billing is $415 per month for Developer instead of $499 and $1,579 per month for Scale instead of $1,899, about 17% off, invoiced once a year. Overage and add-ons are still billed monthly in arrears.
Yes. We sign a BAA on Developer, Scale and Enterprise before you upload any PHI-bearing test case. Inputs, outputs and traces are encrypted with AES-256 at rest and TLS 1.2+ in transit, stored in the region you choose, and never used to train models. The BAA covers GONATURELLE as a testing service; it does not make your production agent HIPAA compliant on its own.
Procurement questions, security reviews or a custom quote: talk to sales or email sales@gonaturelle.si.
“It paid for itself in week two. The gate blocked a prompt change that dropped the member-verification call from our prior-auth agent, two hours before the release train.”
14-day free production trial with 5,000 eval runs and every vertical pack. No card required.