About Horizon
Horizon Analytics Labs, a New York company that goes by Horizon, sells independent audits of AI training datasets. For an agreed sample of a vendor's tasks, it checks the instructions, the reference solution, the scoring tests and the environment each task runs in. It re-runs suspected failures in a clean environment. A private report then lists confirmed defects, their severity and what to fix.
Horizon serves dataset vendors that prepare tasks for AI labs, teams that turn company workflows into training environments, and publishers of benchmarks. Horizon scanned 5,241 tasks across 20 public datasets in an audit published in September 2026 and confirmed 29 broken ones. Jun Cheng Tan is chief executive officer. Horizon's founders have experience at Databento, a market data provider. The team also has experience at CareEvolution, a healthcare data company.
Funding Rounds (SEC Form D)
| First sale | Filed with SEC | Amount sold | Investors | Avg. check | Terms | Filing |
|---|---|---|---|---|---|---|
| Oct 18, 2026 | Oct 27, 2026 | $50,000of $1.0M target | 1 | $50,000 | Options, Simple Agreement for Future Equity (SAFE)Rule 506(b) | Form D ↗ |
First sale is the date the first investor in the offering committed to invest. Companies must file Form D within 15 days of that date, so filing dates run later.
Why Investors Backed Horizon
Horizon sells independent quality checks for the datasets that AI labs use to train and test their models. Buyers can reject broken tasks, and that costs a vendor money when payment depends on acceptance. In a public review of 20 datasets, Horizon confirmed 29 broken tasks. Horizon's founders have experience at Databento, a market data provider.
The problem it targets
Dataset vendors build tasks that AI labs use to train and test models. A buyer may reject a task, sometimes with little explanation. The cost is direct when payment depends on acceptance. Broken tasks also distort test scores. In a research note published September 21, 2026, Horizon pointed to an Epoch AI audit that reviewed 15 benchmarks and labeled nine of them flawed. In its own audit, Horizon confirmed that most of the failures made models look better than they were. In those tasks, the answer was left inside the task, the tests checked only part of the work, or the scoring test rewarded shortcuts.
What the product does
Horizon audits an agreed sample of a vendor's tasks. It compares the instructions, the reference solution, the scoring tests and the environment. Horizon's library has more than 50 checks, and a person verifies important findings before they reach a report. Each private report lists confirmed defects, severity, evidence and what to fix. Horizon can retest fixed tasks. The first consultation is free, and scope and price come in writing before paid work begins. It also publishes a public quality index of datasets and runs a waitlist for AI labs that want to screen tasks before costly model runs.
Progress so far
Horizon has not named an audit client or shared revenue figures. In September 2026 it published an audit of public task datasets. Horizon scanned 5,241 tasks across 20 datasets and confirmed 29 broken ones. Horizon's ranking puts KUMO first, with 14 of 15 reviewed tasks clean. Horizon chose those tasks for deeper review, so the ranking does not show a vendor's overall failure rate. Horizon is a member of NVIDIA's Inception program. It links that to Flur, an AI tool it built. Flur turns floor plans into 3D interiors for hotel and retail teams.
We wrote this summary from public statements by Horizon and its investors. It is not investment advice.
Leadership Team
Executive officers and directors listed on Horizon's Form D filed Oct 27, 2026.
Sources
- SECHorizon Analytics Labs, Inc. filing history
- CompanyCompany page, horizonanalyticslabs.com
- CompanyHow it works, horizonanalyticslabs.com
- CompanyPublic data quality index, horizonanalyticslabs.com
- CompanyDataset audits for vendors, horizonanalyticslabs.com
- CompanyDataset screening for model labs, horizonanalyticslabs.com
- CompanyClient experience page, horizonanalyticslabs.com
- CompanyResearch note of September 21, 2026: public benchmark dataset audit
- CompanyFlur, home page (product built by Horizon), theflur.com