AI tool
Perit Marketplace
Perit Marketplace lets companies and domain experts sell proprietary datasets or build expert AI tasks for frontier AI labs. It is for startups, studios, agencies, and specialists who own real-world data or deep domain knowledge.

About Perit Marketplace
Perit Marketplace is a platform that connects data owners and domain experts with frontier AI labs. Users can submit proprietary datasets-audio, video, code, CAD files, and other modalities-or build reinforcement learning tasks on the platform's workbench for labs to review. The company reports $11M in revenue over 4.5 months and operates as part of the YC F26 batch.
Review
Perit Marketplace functions as a two-sided pipeline. On one side, it accepts data submissions across 29 listed datasets and nine modalities. On the other, it lets experts construct RL tasks with structured quality checks before those tasks reach labs. The platform handles packaging, difficulty calibration, and the transaction layer between contributors and buyers.
Key Features
- Data submission portal. Users can submit audio, video, code, documents, images, CAD, clinical, and enterprise workflow data. The platform currently lists 29 datasets, including 2.89M+ hours of audio, 4.74M+ hours of video, and 21.1M+ lines of code.
- Expert task workbench. Domain experts build agentic tasks and benchmarks inside a hosted environment with sandboxed VMs and model credits. Each task bundle includes a weighted rubric and stays in a private workspace until published.
- Seven quality gates. Tasks must pass Build, Oracle, Do Nothing, New Seed, Difficulty Probe, Human Review, and Publish stages. Only tasks where frontier models score between 10% and 60% pass rate count as valid.
- Difficulty calibration. The system uses OpenAI models and Grok to probe task difficulty, aiming to keep every accepted task within the 10%-60% solve range.
- Auditable runs. Every task execution is isolated, and every score traces back to the exact run that produced it.
Pricing and Value
Pricing for data submissions and task payouts is not yet publicly defined. The platform states that experts get paid when a lab picks their task, and successful contributors can scale to hundreds of tasks per week. No subscription fees, listing fees, or commission percentages are mentioned in the available product information.
Pros
- Accepts a wide range of data modalities, from CAD to clinical records, not just text and images.
- The seven-gate quality system filters out weak tasks before labs ever see them.
- Sandboxed VMs isolate every task run, which creates a clear audit trail for scores.
- Revenue figure of $11M in 4.5 months suggests existing lab relationships and demand.
Cons
- Pricing structure is opaque-contributors can't see what their data or tasks might earn before submitting.
- The platform is not well suited for contributors who lack domain depth or cannot produce structured, rubric-based tasks that survive the seven-gate pipeline.
- The marketplace currently lists only 29 datasets, which limits variety for labs browsing available data.
Perit Marketplace fits contributors who already hold proprietary datasets or deep domain expertise and want a structured path to monetize them with frontier labs. It's less relevant for generalist crowd workers or those expecting open-market pricing controls. The quality gates and calibration requirements mean the barrier to entry for task builders is real, and only work that meets a specific difficulty band moves forward.









