Pricing
Free under 8 GPUs. Per GPU after that.
Priced per managed GPU: a GPU that passed through the admission hook that month. Blocking is never the paywall: the CLI, the PR bot and the hook block on the free tier too. Early-access rates lock in when you subscribe.
Free
- CLI, MCP server and Claude Code plugin
- The 50 structural checks and the PR bot
- The API, at the free quota
- The self-hosted verifier container
- The admission hook, free under 8 GPUs
Change pre-check report
- Your configs and the last 30 days of jobs: what each asked for against what it needed, card by card
- A pre-check of each change in that window: model or checkpoint, training recipe, parallel layout, serving config
- Credited in full toward a subscription
Team
- Admission hook at submit: SLURM
cli_filteror SkyPilot admin policy - The pre-check on every model change PR, in CI
- A ledger of predicted against measured, per job
- Change alerts for the models you run
Per managed GPU, billed annually: $180 per GPU a year, minimum 32 GPUs ($5,760 a year). Teams that only use pull requests can pay $20 per active committer a month instead.
Talk to us →Enterprise / self-hosted
- Everything in Team
- SSO
- The self-hosted container inside your own network, with support, and an offline proof you can run
- Unlimited change history for private repositories
- Your own house rules and limits
Per managed GPU, billed annually: $300 per GPU a year.
Book a call →What each tier gates
Read from the plan cards above; nothing here is in addition to them.
| What | Starts at | Docs |
|---|---|---|
| CLI, MCP server and Claude Code plugin | Free | CLI, hook, MCP |
| The 50 structural checks and the PR bot | Free | Bot |
| The API, at the free quota | Free | API |
| The self-hosted verifier container | Free | Self-hosted |
| The admission hook | Free under 8 GPUs; Team per managed GPU after that | Hooks |
| Your configs and 30 days of jobs, requested against needed, and a pre-check of each change in that window | Report, $3,000 flat, credited in full toward a subscription | |
| The pre-check on every model change PR, in CI | Team | Bot |
| A ledger of predicted against measured, per job | Team | |
| Change alerts for the models you run | Team | Change feeds |
| Blocking a change that will not fit, in the CLI, the PR bot and the hook | Free | Hooks |
| SSO; the self-hosted container with support, and an offline proof; unlimited change history for private repositories; your own house rules and limits | Enterprise / self-hosted | Self-hosted |
The design app
Neurarch app (design canvas and agent). Free with 15 agent messages a day. Pro, $19/mo: 500 agent messages a month, Hugging Face, arXiv and code import, notebook and model-card export. Pro Plus, $39/mo: 2,000 agent and 200 Claude messages, workspaces, live training dashboard. Existing subscriptions are unchanged.
Questions
What counts as a GPU?
A managed GPU: one that passed through the admission hook that month, counted once however many jobs it ran. Under 8, it is free. Team is billed annually with a minimum of 32 GPUs ($5,760 a year).
Do I pay to block a change?
No. Blocking is never the paywall: the CLI, the PR bot and the hook block on the free tier. Team pays for scale past 8 GPUs and the record; Enterprise for SSO, the supported self-hosted container and unlimited private change history.
Is anything billed per seat?
Only if you choose it. Teams that use only the pull-request side can pay $20 per active committer a month instead of per GPU; a team already paying per GPU gets the pull-request side included. The older design app has per-user plans.
What leaves my network?
Depends on where it runs; every surface is listed on the trust page. The self-hosted container makes no outbound connection.