Comparison · Verified May 2026
Looking for a Gretel.ai alternative?
Gretel.ai was acquired by NVIDIA in March 2025. The standalone gretel.ai SaaS is no longer operating; the technology now lives inside NVIDIA NeMo. Here is what changed, and where SynthForge fits versus the NeMo-rebranded products.
TL;DR
Gretel.ai's standalone product is gone. gretel.ai now redirects to NVIDIA, and the gretelai GitHub org was archived on 2026-02-18. The capabilities split into two NeMo microservices: Data Designer (schema-driven generation, with optional LLM-generated columns) and Safe Synthesizer (differentially-private synthetic data trained on a real seed dataset). If you used Gretel because you had a real dataset and needed privacy-preserving copies, NVIDIA NeMo Safe Synthesizer is the successor. If you used Gretel for greenfield schema-driven test data, SynthForge is a closer match: free, web-based, no GPU, and no enterprise contract.
Recent context
TechCrunch reported the NVIDIA acquisition on 2026-03-19. As of May 2026: gretel.ai 301-redirects to nvidia.com; the gretelai GitHub organization (including gretel-synthetics) is archived; gretel.ai/pricing returns 404. The legacy Gretel free tier (15 credits/mo, ~100K records, $2/credit overage) is no longer the live pricing model. NVIDIA NeMo Microservices are licensed via NVIDIA AI Enterprise (sales-gated, GPU-shaped pricing).
When NVIDIA NeMo (the Gretel successor) is the right call
- • You have a real sensitive dataset and you need a privacy-preserving synthetic copy with differential-privacy guarantees. NeMo Safe Synthesizer is the productized version of the Gretel Tabular DP-SGD workflow.
- • You are already on NVIDIA AI Enterprise or running NIM-based infrastructure, so NeMo microservices fit your existing stack.
- • You need to integrate synthetic-data generation into a larger NeMo / agentic-AI pipeline (data designer + RAG + fine-tuning).
- • You need PII detection / classification driven by the Nemotron model family.
- • You can wait hours, not seconds, for output (DP-SGD training time is real) and you have a meaningful seed dataset to train on.
When SynthForge is the right call
- • You do not have a real dataset to train on. You are pre-launch, building a schema, or generating test data for a new feature. NeMo Safe Synthesizer cannot help you here; SynthForge can.
- • You want a free, web-based product with no enterprise contract, no GPU, and no AI Enterprise license.
- • You need output in seconds, not hours: SynthForge is parametric / sampler-driven, not training-driven.
- • You need direct multi-dialect SQL output (Postgres, MySQL, SQLite, SQL Server, MariaDB, DuckDB, CockroachDB). NeMo emits CSV/JSON/Parquet via API; dialect-aware DDL is not the focus.
- • You want pre-built ML training dataset templates with baseline model evaluation, not a microservice-orchestration framework.
Feature comparison
Verified against primary sources in May 2026.
| Feature | SynthForge | Gretel.ai (now NVIDIA NeMo) |
|---|---|---|
| Product status (May 2026) | Live, hosted, free | Gretel SaaS shut down. Capabilities live in NVIDIA NeMo Microservices. |
| Requires a real seed dataset | No | Safe Synthesizer: yes (DP-SGD trains on it). Data Designer: optional. |
| Differential privacy | No | Yes. Safe Synthesizer with configurable epsilon, group-level / user-level privacy |
| Time to first output | Seconds | Data Designer: minutes. Safe Synthesizer: hours (training-bound). |
| Hosting model | Hosted SaaS, free | NVIDIA AI Enterprise license (sales-gated, GPU-shaped pricing). gretel.ai SaaS is gone. |
| Free tier | Yes | Legacy Gretel free tier (15 credits/mo) no longer applies. Current NeMo free-tier status: unverified; assume enterprise-gated. |
| Multi-table FK preservation | Yes (single-column, by construction) | Legacy Gretel Relational supported it. Under the NeMo rebrand: capability appears regressed or not yet exposed (unverified). |
| Output formats | CSV, SQL (7 dialects), JSON, JSONL, Parquet | CSV, JSON, Parquet via NeMo API. SQL dialect-aware output: not a focus. |
| Privacy report / metrics | Privacy benchmarks (DCR, NNDR) for evaluation | Synthetic Quality Score (SQS) and Data Privacy Score (DPS) on Safe Synthesizer output |
| ML training dataset templates | Yes (6+ pre-built domain templates with baseline evaluation) | Not a packaged feature; NeMo is an orchestration layer |
| PII detection | No | Yes. Nemotron-PII model lineage |
| Open source | Not OSS today | Legacy gretel-synthetics archived 2026-02-18. Active OSS is now github.com/NVIDIA-NeMo/DataDesigner. |
Pricing comparison
SynthForge
All features. No credit card. Per-account rate limits and a 10M-row hard cap per generation request.
Gretel.ai (now NVIDIA NeMo)
The gretel.ai/pricing page no longer resolves. Free tier (15 credits/mo) and $2/credit overage are no longer in effect.
Sales-gated, GPU/subscription based via NVIDIA AI Enterprise. Public list pricing not surfaced. Treat as enterprise procurement.
What SynthForge does not do that Gretel.ai (now NVIDIA NeMo) does
Honest tradeoffs, in case they decide the comparison for you.
- • If you need differential privacy, SynthForge does not have it. NeMo Safe Synthesizer does. This is the cleanest case for picking NeMo over SynthForge.
- • If you have a real sensitive dataset and you need to train a model on it for a privacy-preserving synthetic copy, SynthForge is not designed for that workflow.
- • If you are building agentic-AI pipelines on NVIDIA NIM, NeMo's ecosystem fit is a real advantage SynthForge cannot match.
- • NeMo's PII detection (Nemotron lineage) is meaningfully ahead of anything SynthForge offers.
Frequently asked questions
Is Gretel.ai still available?
What replaced Gretel?
Is gretel-synthetics still maintained?
Can SynthForge generate differentially-private synthetic data?
Why do people search for Gretel alternatives now?
If I want a privacy-preserving synthetic copy of my prod DB, what should I use?
Other SynthForge comparisons
Try SynthForge for free
Design a multi-table schema, generate referentially-intact data, and export to your database. No credit card.