Gretel Alternative: Gretel Was Acquired by NVIDIA. Here Is What Changed.
Gretel.ai is no longer a standalone product. The domain redirects to NVIDIA, the pricing page returns a 404, and the flagship open-source repository is archived. The successor, NVIDIA NeMo Data Designer and Safe Synthesizer, requires Docker or Kubernetes and an NVIDIA NGC account. This page shows what happened, what changed, and what to use now.
Last reviewed: 2026-08-01
What the product is
Gretel.ai was a synthetic-data platform later acquired by NVIDIA. Product surfaces, pricing pages, and API availability changed after the acquisition. This page states what was verified as of 2026-08-01 and marks unpublished items explicitly rather than inventing continuity.
Comparison
| Capability | Gretel | Generate-Data |
|---|---|---|
| Company status | Acquired by NVIDIA (verify current product home as of 2026-08-01) | Independent Generate-Data.com product |
| Primary job | Privacy-preserving synthetic data / ML workflows (historical positioning) | Self-serve test-data schema builder with duplicate generation |
| Free path | Not published as of 2026-08-01 for post-acquisition packaging | Anonymous free path with stated caps |
| Duplicate/fuzzy pair generation | Not published as a first-class QA answer-key feature as of 2026-08-01 | Master ID / Duplicate Type columns |
| Export formats | Not fully republished as of 2026-08-01 | CSV, JSON, XML, Parquet, XLSX, JSONL, hf-datasets |
| Signup / sales motion | Enterprise-oriented historically; current motion: verify on NVIDIA/Gretel properties | Self-serve web; anonymous path available |
Updated 2026-08-06
The labeled ground-truth difference
Most tool-vs-tool comparisons on this site are about schema builders and export formats. This one is about what happens after the export: can the match rules being tested actually be scored.
A Known-Duplicate Pair File is a synthetic dataset where every duplicate pair is tagged with its Master ID and Duplicate Type, so match-rule testing has a documented answer key instead of an assumed one. Synthetic data notice: the file is fabricated data, not real records. Generate-Data ships this output, already recorded in the "Duplicate/fuzzy pair generation" row of the comparison table above. Whether a given alternative tool ships an equivalent answer key is stated, where sourced, in that same row; where it is not published, the table says so and dates the check rather than guessing.
Settle it on your own data
If the comparison above is close for your use case, settle it on your own data: request a free, no-call Match-Ready Sample shaped to your own domain (patient, supplier, B2C, or another), with the Known-Duplicate Pair File answer key attached, and run your own match rules against the Master ID / Duplicate Type labels before deciding.
Get a Match-Ready SampleWhere Gretel is stronger
Legacy Gretel strength was privacy-preserving synthetic data for ML and analytics workflows, including differential-privacy oriented tooling. Treat any remaining Gretel Cloud claims as time-sensitive after the NVIDIA acquisition.
Where Generate-Data is the better fit
Generate-Data fits teams that need an instant browser schema builder, labeled duplicates for match/merge QA, and file exports (including JSONL / HuggingFace) without an enterprise synthetic-data platform purchase.
Pricing (dated 2026-08-01)
| Plan | Published price | Key inclusions / notes |
|---|---|---|
| Post-acquisition packaging | Not published as of 2026-08-01 | Confirm on current NVIDIA/Gretel properties before quoting |
| Generate-Data anonymous | Free within caps | 100 rows, 6 fields, 3 exports, CSV only |
| Generate-Data signed-in | Free tier with expanded exports | See product limits in-app |
Generate a dataset
Open the free generator to build a custom schema. Anonymous use needs no account within the stated caps.
Frequently asked questions
Is Gretel.ai still available?
Not as a standalone product. As of 2026-08-01, `gretel.ai` redirects to NVIDIA, `gretel.ai/pricing` returns a 404, and the `gretelai/gretel-synthetics` GitHub repository is archived with `"archived": true` per the GitHub API (last push 2025-06-24). [Source: `competitors/gretel.md` §1, curl and GitHub API checks, 2026-08-01] (Question framing observed on `synthforge.io/alternatives/gretel/`, 2026-08-01.)
Did Gretel shut down?
The company was acquired by NVIDIA, not shuttered. CRN reported the acquisition, citing Wired, and stated that the exact financial terms were not confirmed. Do not print a deal value on the page. The exact acquisition date is reported inconsistently across sources (one says March 2025, another says March 2026 on the same page). What is consistent: the product was absorbed into NVIDIA's NeMo platform and is no longer sold or documented as a standalone product. [Source: `competitors/gretel.md` §1, 2026-08-01]
What replaced Gretel?
NVIDIA NeMo Data Designer and NVIDIA NeMo Safe Synthesizer (Early Access) are the successor microservices. They require Docker or Kubernetes and an NVIDIA NGC early-access account. They are designed for training LLMs and agentic AI systems, not for generating QA test fixtures. [Source: `competitors/gretel.md` §2, §5, 2026-08-01]
Can I still use the Gretel open-source library?
The `gretelai/gretel-synthetics` repository is archived and read-only as of the GitHub API check on 2026-08-01 (last pushed 2025-06-24). Archived repositories remain installable but receive no maintenance. Whether the library still functions against its dependencies was not tested and is not claimed here. [Source: `https://api.github.com/repos/gretelai/gretel-synthetics`, 2026-08-01]
Do I need a GPU to use the Gretel successor?
GPU configuration is a documented setup step for the NeMo Microservices ("Quickstart Installation (GPU configuration and SDK setup)"). [Source: `https://docs.nvidia.com/nemo/microservices/latest/about/index.html`, 2026-08-01]
What is a Known-Duplicate Pair File and why does it matter for match-rule testing?
A Known-Duplicate Pair File is a synthetic dataset where every duplicate pair is tagged with a shared Master ID and a Duplicate Type describing how the rows vary (exact copy, typo, nickname, and similar). It matters for match-rule testing because it gives you a documented answer key: which rows should match is stated, not inferred from eyeballing the export.
Does a generic synthetic-data export include a duplicate answer key?
It depends on the tool. Some synthetic-data exports are records only, with no indication of which rows are intentional duplicates of the same entity. Generate-Data ships a Known-Duplicate Pair File with Master ID and Duplicate Type columns; whether a given alternative publishes an equivalent answer key is noted in the comparison table above, sourced where available.
How do I get a labeled sample instead of just a synthetic export?
Request a Match-Ready Sample: a free, no-call sample shaped to a domain you name (patient, supplier, B2C, or another), delivered with the Master ID / Duplicate Type answer key built in. It is the same mechanism behind the Known-Duplicate Pair File, sized for evaluation rather than a full engagement.