Sandbox
$0
free forever
for a single prompt engineer testing the notebook.
- →Unlimited cells, unlimited drafts
- →Single seat · single project
- →Per-token cost subscripts (4 providers)
- →Eval gate disabled (no promotes)
- →Community Discord
tokencraft · pricing · per promoted version
You pay when a prompt ships — not when a developer opens the tab.
No per-seat extortion. No usage caps that triple your invoice when you actually ship. One number: $0.05 for every prompt version that clears the eval gate and gets promoted to HEAD. Drafts, dry-runs, and rollbacks are free.
Lab
$0.05
per promoted version
the honest wedge — you pay only when a version ships.
- →K1.8 · semver promote with eval gate
- →K1.4 · token-budget enforcer (mid-stream halt)
- →K6.2 · OTLP traces to Honeycomb / Grafana
- →Up to 5 seats · unlimited projects
- →Email support · 24h response p50
Foundry
talk to us
annual contract
for teams running >1K promotes/mo with SSO + SLA + audit.
- →Everything in Lab
- →SAML SSO · SCIM provisioning
- →Custom eval gates + private eval sets
- →Self-hosted OTel collector option
- →99.9% SLA · named CSM
What counts as a “promoted version”?
A semver bump that clears the eval gate (HumanEval ≥ HEAD − 5%) and lands on HEAD. Drafts, dry-runs, rolled-back versions, and override-rejected promotes do not count.
Is there a usage cap?
No. The Lab tier is pure per-promote billing. Run 10 cells/day or 10,000 — the invoice only tracks promotions that shipped.
Do you charge for eval runs?
No. Cell runs, eval-set executions, and the right-rail provider projections are all free at the Lab tier. We charge when you commit a release.
Can I bring my own evals?
At Lab, yes — upload any JSONL with prompt+expected. At Foundry, the eval gate is fully configurable, including private benchmark sets behind your own auth.