Cerebrium

What it does

Cerebrium is a serverless GPU platform for deploying voice agents, video models, LLMs and other AI workloads. It runs your existing code or Dockerfile without rewrites, with autoscaling and per-second billing. The platform offers multi-region deployments, observability and security features for production AI.

Cerebrium is a serverless GPU platform for running AI workloads such as voice agents, video models and LLMs. You point it at your entry point or Dockerfile, and it runs your application as is, with autoscaling and per-second billing. The homepage states that containers launch in seconds with memory and GPU snapshotting for fast restores, and that the platform handles bursts and scale-outs automatically. It lists multi-region deployments, OpenTelemetry integration, WebSocket endpoints, REST API endpoints and custom Dockerfiles among its features. Security information mentions SOC 2, HIPAA, GDPR and ISO standards, data residency and gVisor isolation. Pricing is based on actual compute time measured in seconds, with a Hobby plan described as Free plus compute per month. A careful user should check region availability, concurrency limits per plan and whether the snapshotting approach fits their model startup pattern. Costs depend on GPU choice and runtime.