Modal
Serverless AI cloud for inference, training, batch jobs and scalable sandboxes that give agents their own computers
Modal is a serverless cloud for developing, training and serving AI applications, and its Sandboxes product gives agents their own computers. Developers define containers and functions in Python code, and Modal handles provisioning, GPU capacity, scaling and isolation for everything from coding agents to RL rollouts.
Key features:
- Create millions of concurrent sandboxes programmatically in under a minute.
- Run any custom image and snapshot and restore sandboxes for fast startup.
- Burst CPU and memory on demand without over-allocating.
- Serve models through OpenAI-compatible endpoints or custom inference stacks.
- Fine-tune on single or multi-node GPUs with RL, SFT and LoRA.
- Fan out parallel functions, batch jobs and web endpoints that scale to zero.
Modal bills per second for CPU, GPU and memory, with sandboxes at non-preemptible rates. The Starter plan is $0 plus compute with $30 of free credits a month; Team is $250/month plus compute with $100 in credits, and Enterprise adds Okta SSO, audit logs and HIPAA.