Beta9

Beta9 is the open source runtime behind Beam, a serverless platform for running AI workloads on GPUs and CPUs without managing servers. Developers decorate ordinary Python functions to turn them into autoscaling inference endpoints, task queues or isolated sandboxes, then ship them with a single deploy command. The runtime uses a custom container scheduler, embedded caching and memory snapshots so containers boot in under a second, and it fans work out across hundreds of containers in parallel. It is aimed at engineering teams shipping model inference, agents and batch jobs. SDKs cover Python, TypeScript and Go. Workloads can run on Beam's own cloud or on the customer's AWS, GCP or bare metal accounts, with bursting across clouds and more than 30 regions. Available hardware includes RTX 4090, RTX 5090, A100, H100, H200 and B200 GPUs. Sandboxes give an isolated environment for running untrusted code at scale. Serverless usage is billed by the millisecond only while code runs. Published GPU rates include $0.000192 per second for an RTX 4090, $0.000303 per second for an RTX 5090 and $0.000972 per second for an H100, with CPU and memory charged separately. Sandboxes start from $0.319 per hour for one core and 8 GiB. On demand machines are priced per hour, from $1.36 for an A100, $1.83 for an H100, $2.09 for an H200 and $4.11 for a B200. Usage tiers are Developer at $0 per month plus usage with one seat, Team at $89 per month plus usage with three seats and $25 per extra seat, and Growth on a contact basis. The runtime is licensed under the AGPL and can be self hosted; Beam is the managed service built on it.