Gretel.ai vs Mockaroo

Quote on request

From $60 per year (indicative, Silver)

The old credit-based self-serve tier is gone. The successor microservices are licensed through NVIDIA AI Enterprise; NeMo Data Designer can be tried on build.nvidia.com.

Cheap for what it does; higher tiers raise row limits and add API quota.

Free for up to 1,000 rows per download and a limited number of saved schemas.

Emerging

NVIDIA (formerly Gretel) · since 2019

Established

Mockaroo LLC · since 2014

• Formal differential-privacy guarantees rather than best-effort masking, now via DP-SGD in NeMo Safe Synthesizer.
• Synthetic data preserves statistical relationships useful for analytics and model training.
• Backed by NVIDIA's platform investment rather than a single startup's runway.

• No code required — an analyst or tester can design a dataset themselves.
• Very large library of realistic field types including regional formats.
• Formula fields allow simple business rules between columns.

• No longer buyable as Gretel — you are procuring NVIDIA AI Enterprise, which is far heavier than the old self-serve signup.
• Model training adds time and cost compared to rule-based masking.
• Safe Synthesizer trains on a real dataset, so it does not help with greenfield data that does not exist yet.

• Referential integrity across related tables is limited.
• Free-tier row limits are low for realistic volume testing.
• Data is generated on a third-party service, which needs a privacy check for schema sensitivity.

Regulated enterprises already standardising on NVIDIA AI Enterprise. Teams that wanted Gretel's self-serve tier should look at Tonic or Neosync instead.

Quickly producing a realistic dataset for a demo, a load test or a new environment.