Product GuidesstableUpdated 2026-05-15

Optimizations — Sandbox

Simulate cheaper or better model alternatives.

What this page is for

Sandbox compares the current model workload with compatible alternatives using usage, pricing, quality, latency, token behavior, and workload constraints.

Who uses it

  • Platform teams
  • FinOps teams
  • ML/AI engineers

What you see

  • Current cost.
  • Alternative models.
  • Recommended alternative.
  • Cheapest alternative.
  • Excluded alternatives with reasons.
  • Confidence and estimation mode.

What you can do

  • Run scenario.
  • Compare alternatives.
  • Create recommendation from scenario.

Permissions and safety

  • Sandbox access may be plan-gated.

Operational notes

  • The simulation should not recommend incompatible models just because they are cheaper.

What this page is not for

  • Applying changes directly in production.
  • Replacing evaluation pipelines.
  1. Confirm that you are in the right workspace and environment.
  2. Use filters before changing configuration or interpreting results.
  3. Open the detail drawer when available.
  4. Check related objects such as alerts, approvals, costs, policies, runs, deliveries, or audit entries.
  5. Apply changes only when the impact is clear.
  6. Re-check the page after the backend refreshes.
  7. Export or copy IDs when you need to escalate.