Free
For individuals and side projects.
$0 / month
- 1 user, unlimited notebooks and runs
- Experiment tracking and comparison
- Shared CPU runtimes, 20 hours / month
- 5 GB artifact storage
- 1 deployed endpoint
- Community support
Every plan includes unlimited runs, models and notebooks. You pay for the people who build, plus the compute they actually use, billed by the minute.
For individuals and side projects.
For teams shipping models together.
For large ML teams and regulated industries.
| Feature | Free | Team | Enterprise |
|---|---|---|---|
| Notebooks & compute | |||
| Jupyter-compatible notebooks | ✓ | ✓ | ✓ |
| Real-time co-editing and comments | — | ✓ | ✓ |
| CPU runtimes | 20 h / month shared | Metered | Metered or your own |
| GPU runtimes | — | Metered | Metered or your own |
| Bring your own compute | — | — | ✓ |
| Experiments & data | |||
| Automatic experiment tracking | ✓ | ✓ | ✓ |
| Run comparison and sweeps | ✓ | ✓ | ✓ |
| Artifact storage | 5 GB | 100 GB / seat | Custom |
| Dataset versioning and lineage | — | ✓ | ✓ |
| Data source connectors | Files only | All | All, plus private connectors |
| Models & deployment | |||
| Model registry | Basic | ✓ | ✓ |
| Approval workflows | — | ✓ | ✓ |
| Deployed endpoints | 1 | Unlimited | Unlimited |
| Batch jobs | — | ✓ | ✓ |
| Canary, shadow and rollback | — | ✓ | ✓ |
| Deploy into your VPC | — | ✓ | ✓ |
| Monitoring & assistant | |||
| Drift and data quality monitoring | — | ✓ | ✓ |
| Slack, Teams and PagerDuty alerts | — | ✓ | ✓ |
| AI assistant | Limited | ✓ | ✓ with BYO keys |
| Security & support | |||
| Role-based access control | — | ✓ | ✓ |
| SSO (SAML, OIDC) and SCIM | — | — | ✓ |
| Audit log | — | 90 days | Custom, exportable |
| Self-hosted / single-tenant | — | — | ✓ |
| Data residency | — | EU or US | Any region |
| Support | Community | Email, next business day | 24/7 with SLA |
Anyone who can open or edit a notebook, or register or deploy a model. People who only view dashboards, receive alerts or approve models do not need a seat. Service accounts used by CI are free.
GPU and CPU machines are billed per minute while attached to a notebook or a training job, at published rates per machine type. Idle machines shut down automatically after a period you set, and each project can have a monthly cap. Bringing your own GPUs on Enterprise removes the metered charge entirely.
Yes. Everything exports completely: notebooks, training runs with their metrics, saved models, dataset versions and approval history, and imports into a self-hosted deployment. Notebooks do not change.
No. Your notebooks, training data, runs and models are never used to train anything, including our assistant. The assistant reads your experiments to answer your questions and can run on your own model provider keys on Enterprise.
Prediction APIs keep answering requests through the end of the billing period, and you can export every model's weights and code in standard formats at any time. We will not hold your models hostage.
Yes. Accredited institutions and registered non-profits get Team features at a reduced per-seat rate, with instructor tooling for courses. Ask through the contact form.
Team is available monthly or annually with a discount for annual. Enterprise is annual.
Tell us how many people train models, what they train, and where predictions run today. We will recommend a plan and set up a trial with your own notebook in it.