Skip to content
Sandworld

Pricing

Start free. Pay for verified work.

Indicative prices. Every engagement is scoped and fixed in writing before work starts.

Sandbox

Free

For trying the worlds

  • All public worlds and tasks
  • Public API, rate-limited
  • Verifier QA reports
  • Tool definitions for Claude, OpenAI and MCP
Open the worlds

Pilot

Most teams start here

from $25k

4–6 weeks, fixed scope

  • One custom twin of your target app category
  • 200 verified tasks with a QA pack per task
  • Baselines plus your model's results
  • Expert demonstrations recorded in the twin
  • Delivery in your harness's format
Book a pilot

Environment licence

from $20k / twin / yr

Non-exclusive

  • Full task suite and seeded variants
  • Quarterly refresh with held-out tasks
  • Private API keys and episode logs
  • Exclusive builds priced at 4–5×
Talk to us

Expert data

Priced per accepted task.

Rates vary with the expertise and time a task needs. Experts see their pay before they start; you see quality numbers before you pay.

Preference judgmentsfrom $6 per accepted pair
Rubric gradesfrom $12 per item
Agent demonstrationsfrom $15 per verified episode
SFT answersfrom $25 per answer
Red-team findingsfrom $20 per reproducible finding

FAQ

Pricing questions

Are these prices final?

They're indicative starting points. Final pricing depends on domain, volume, turnaround and exclusivity, and is fixed in the statement of work before any work starts.

What's the difference between exclusive and non-exclusive?

Non-exclusive environments and datasets may also be licensed to others. Exclusive builds are yours alone, priced at roughly 4–5× the non-exclusive rate.

Do you charge for rejected work?

No. Expert data is billed per accepted task; rejected and unresolved items are not billed.