Mercor is one of the largest expert-data companies, with a vetted network of professionals, benchmark work through APEX and RL environments from its acquisition of Deeptune. Sandworld is an early-stage company building deterministic app twins with published verifier QA, and its expert program is just starting. If you need many vetted professionals quickly, Mercor is the established option; Sandworld is a narrower bet on verified app environments and the demonstrations recorded inside them.
How do Sandworld and Mercor compare at a glance?
| Mercor | Sandworld | |
|---|---|---|
| Company stage | Established; reportedly a $2B gross run-rate in June 2026 | Early-stage; no customers yet |
| Expert network | Large vetted professional network; average contracted rate of $107/h | Program just starting; no network yet |
| Vetting | Assessments, AI video interview, ID check with selfie, optional background check | Assessments and gold tasks, as the program opens |
| RL environments | Hundreds of recreated enterprise applications, via Deeptune | Four deterministic app twins with seeded variants |
| Benchmarks | APEX, including APEX-Agents (480 tasks) | Computed baselines and verifier QA for each world |
| Security | Okta with MFA; an enterprise browser on some projects | No certifications yet; SOC 2 readiness is planned |
What does Mercor offer?
- Expert data and data licensing from a network of professionals, at an average contracted rate of $107 an hour in October 2026 (Mercor).
- RL environments. Mercor agreed to acquire Deeptune, which had “recreated hundreds of enterprise applications”, and defines an environment as software plus tasks plus verifiers (announcement).
- Benchmarks. APEX-Agents has 480 tasks, and its top score at launch was 24.0%; benchmark partners include Harvey, Ramp, Cognition and Sierra (APEX).
- A documented vetting pipeline: assessments that prohibit LLMs and disable copy-paste, a roughly 20-minute AI video interview, an ID check with a selfie through Persona, an optional background check and Stripe payouts.
- Quality and security operations: a writer → reviewer → QA → QC pipeline, Okta with MFA and an Island enterprise browser on some projects, and anonymous peer endorsements that feed a reputation score.
Where is Mercor stronger?
- Network and speed. Mercor can staff vetted professionals across many fields now; Sandworld has no network yet.
- Track record. Mercor reportedly reached a $2B gross run-rate in June 2026 and counts OpenAI, Anthropic, Meta and Google among its customers (AI Weekly); it was valued at $10B in October 2025 (TechCrunch).
- Environment breadth. Hundreds of enterprise application recreations, against Sandworld’s six app twins and one cross-app world.
- Mature operations for identity, review and secure work environments.
How is Sandworld different?
- Verifiers you can inspect and rerun. Every Sandworld task publishes the results of five seeded probes — reference, do-nothing, random, truncated and noisy — with an A–C grade, on each world’s page.
- Demonstrations checked by code first. Experts work inside the same twins agents use, and the task’s verifier checks every recording before a person reviews it.
- Mobile-style consumer twins. Quickbasket (grocery delivery) and Coinpurse (a payments wallet) are phone-app twins. They run in the browser today, with a tool surface for agents and a phone UI for people.
- Contributor terms set up front. Each project shows its pay per task before an expert accepts it, and accepted work goes to a visible earnings ledger. Because the program is new, we can’t yet offer the volume of work an established network can.
Which should you choose?
- Choose Mercor if you need many vetted experts quickly across domains, replicas of specific enterprise applications, or a vendor with a long list of frontier-lab customers.
- Consider Sandworld if your priority is environment tasks with transparent, reproducible grading and demonstrations verified inside those environments, and you’re open to piloting with an early-stage team.
For background, see how to buy expert training data and what an RL environment is.