AWS makes Reflection Beam available through SageMaker JumpStart

AWS has made Reflection AI’s Beam coding model available through Amazon SageMaker JumpStart, giving customers a managed catalog route for deploying the open-weight model on AWS.

The JumpStart listing is available in US East (N. Virginia) and US East (Ohio). AWS also outlined SageMaker HyperPod Inference options for teams that need more control over the serving runtime and distributed GPU infrastructure.

AWS lays out three deployment paths. The first is a JumpStart real-time endpoint, which packages the model, serving container, endpoint and invocation API without requiring customers to operate Kubernetes.

The second uses JumpStart to deploy its AWS-managed container and model configuration on a HyperPod cluster orchestrated by Amazon EKS. The third lets teams set runtime arguments, quantization or distributed execution themselves with custom SGLang or vLLM runtimes on HyperPod. AWS’s HyperPod deployment documentation describes the underlying deployment workflow.

For context on the managed catalog route, AWS previously added Meta’s Muse Spark Open to SageMaker JumpStart.

This AWS availability update is distinct from Reflection AI’s broader model launch. Reflection introduced Beam on October 5, describing it as a sparse mixture-of-experts model with 501 billion total parameters and 23 billion active parameters for coding, reasoning and agentic workloads.

AWS says Beam 1.0 comes in BF16, FP8 and NVFP4 checkpoint variants and requires a multi-GPU deployment. Its published configurations map FP8 checkpoints to supported SageMaker instances using NVIDIA H100 or H200 GPUs and the NVFP4 checkpoint to a supported instance using NVIDIA B200 GPUs.

AWS did not provide deployment pricing or a timetable for availability in additional regions.