One subscription; compute is pass-through at cost, never marked up. Seats are free and public runs are unlimited at every tier. The founding cohort locks 50% off for life, and every price below already reflects it.
| Compare plans | Free $0forever Public runs, on us Start free | Entry $49$25/mo billed monthly Start Entry | Most popular Growth $199$100/mo billed monthly Start Growth | Scale $499$250/mo billed monthly Start Scale |
|---|---|---|---|---|
| Best for | Try it | Solo builder | Small team | Company |
| Run existing benchmarks | ✓ | ✓ | ✓ | ✓ |
| Build benchmarks | ✗ | ✓ | ✓ | ✓ |
| Private benchmarks | 0 | 3 | 15 | Unlimited |
| Operators | 0 | 1 | 3 | 10 |
| Concurrent runs | 1 | 3 | 10 | 30 |
| Public runs | Fair use | Unlimited | Unlimited | Unlimited |
| Private retention | n/a | 30 days | 90 days | 1 year |
| Private storage | n/a | 5 GB | 50 GB | 250 GB |
| Queue priority | Background | Standard | Priority | High |
| API rate | 60/min | 300/min | 600/min | 1,200/min |
| Seats | Unlimited | Unlimited | Unlimited | Unlimited |
| SSO | ✗ | ✗ | ✗ | ✓ |
| Onboarding credit | none | $10 | $25 | Committed |
Try the platform on public benchmarks.
Solo builder shipping private evals.
Small team with operators and priority queue.
Company volume, high priority, SSO.
Same limits as the desktop comparison
Subscriptions and licenses are the only thing we charge for. Sandbox time and LLM inference bill from your prepaid balance with no markup and no monthly run quota. Retention and storage limits apply to private runs only; public runs are unlimited and land in the shared registry.
When you outgrow Scale, or need isolation and compliance, we run your own single-tenant stack: managed (we operate it, infra at cost plus a fee) or self-hosted (you run it under an annual license). Talk to us about dedicated capacity.
No. Sandbox time and model inference bill at exact provider cost, pass-through, with no markup. Subscriptions and licenses are the only thing we charge for. Credits are prepaid and runs are hard-capped against your balance, so a run can never spend money you have not put down.
The first cohort of customers locks 50% off for life. The discount applies to the subscription and holds for as long as you stay; prices return to full once the founding window closes. Pay annually to save another 20% on top.
No. We never train on your private data. We only ever use public runs (the ones that land in the shared registry) for analysis and aggregate statistics. Private benchmarks and private runs are yours alone and are never used to train anything.
Public runs land in the shared registry, so everyone benefits from the results; they are unlimited and never storage-billed. Private runs and private benchmarks are yours alone, and how many you can hold plus how long they are retained scales with your tier.
Bring your own model keys any time to bill inference directly. Bringing your own compute means running the whole stack yourself, which is the self-hosted Dedicated option under a license.
You move to Dedicated: your own single-tenant stack, off the shared platform. Run it managed (we operate it, infrastructure at cost plus a management fee) or self-hosted (you operate it under an annual license). This is also the path for data isolation and compliance.
Scores and summaries are kept forever on every tier, so your history never disappears. Public run detail is kept permanently. Private run detail is kept within your tier's private retention window, then deleted.
Yes, upgrade, downgrade, or cancel any time. Because compute is a separate prepaid balance, changing your subscription never puts a run at risk mid-flight.