AWS Adds Reflection AI's Open-Weight Beam Model for Agentic Coding
AWS detailed how developers can deploy Reflection AI's open-weight Beam model for agentic coding through Amazon SageMaker JumpStart and HyperPod Inference.
Quick answer
What did AWS announce about Reflection AI's Beam model?
AWS published a technical guide for deploying Reflection AI's Beam, an open-weight mixture-of-experts model for agentic coding, on Amazon SageMaker. Beam has 501 billion total parameters, 23 billion active parameters, and a reported 1 million token context length, with weights released on Hugging Face under Apache 2.0.
Key takeaways
- Reflection AI released Beam, an open-weight model built for agentic coding tasks including code maintenance, tool use, and multi-step workflows.
- Beam is a sparse mixture-of-experts model with 501 billion total parameters and 23 billion active parameters, and Reflection reports an effective context length of 1 million tokens.
- AWS says Beam's weights are available on Hugging Face under the Apache 2.0 license in BF16, FP8, and NVFP4 checkpoint variants.
- AWS describes two deployment paths: a managed SageMaker JumpStart real-time endpoint, or SageMaker HyperPod Inference with JumpStart, SGLang, or vLLM for Kubernetes-native control over distributed GPU infrastructure.
- At launch, AWS says Beam is available in SageMaker JumpStart in the US East (N. Virginia) and US East (Ohio) regions.
Reflection AI has released Beam, an open-weight model built for agentic coding and long-horizon tasks such as maintaining code, using tools, interpreting results, and completing multi-step workflows. AWS detailed the release in a blog post explaining how developers can deploy the model on its infrastructure.
According to AWS, Reflection AI's mission is to make intelligence open and accessible through open model weights, published research, and open source software, and Beam extends that approach to software engineering and agentic systems.
What Beam is
AWS describes Beam as a sparse mixture-of-experts model for resolving software issues, modifying code, calling tools, and acting on intermediate results. It has 501 billion total parameters and 23 billion active parameters. Reflection reports an effective context length of 1 million tokens.
Reflection has released Beam's weights on Hugging Face under the Apache 2.0 license. Beam 1.0 is available in three checkpoint variants: BF16, FP8, and NVFP4, which AWS says provide different memory footprints and hardware compatibility.
How to deploy it on AWS
AWS lays out two main deployment paths for Beam:
- Amazon SageMaker JumpStart real-time endpoint — a fully managed path that packages the model, container, and endpoint without requiring customers to operate Kubernetes.
- Amazon SageMaker HyperPod Inference — paired with JumpStart, SGLang, or vLLM, for customers who need Kubernetes-native control over distributed GPU infrastructure, serving images, runtime arguments, quantization, or distributed execution.
AWS says Beam can be sourced through SageMaker JumpStart, the Hugging Face Hub, or an optional Amazon S3 mirror.
Hardware requirements
AWS says Beam's 501 billion total parameters require a multi-GPU deployment. For the BF16 checkpoint, which has the largest weight-memory requirement, AWS recommends planning a multi-node configuration with sufficient aggregate GPU memory. For FP8, AWS specifies using ml.p5.48xlarge instances with NVIDIA H100 GPUs, or ml.p5e.48xlarge and ml.p5en.48xlarge instances with H200 GPUs. For NVFP4, AWS specifies ml.p6-b200.48xlarge instances with NVIDIA B200 GPUs.
AWS lists several prerequisites for deployment, including an AWS account with appropriate IAM, networking, and storage permissions, access to Beam through SageMaker JumpStart or Hugging Face, acceptance of model terms, the AWS CLI, Python 3.11 or later, and GPU quota and capacity for a supported configuration.
Availability
At launch, AWS says Beam is available in Amazon SageMaker JumpStart in the US East (N. Virginia) and US East (Ohio) regions.
Source: AWS Machine Learning Blog, "Introducing Reflection Beam: an open-weight model for agentic coding on AWS," published October 16, 2026.
Frequently asked questions
- What is Reflection Beam?
- Beam is an open-weight sparse mixture-of-experts model from Reflection AI designed for agentic coding tasks such as resolving software issues, modifying code, and calling tools, according to AWS.
- How large is the Beam model?
- AWS states Beam has 501 billion total parameters and 23 billion active parameters, with a reported effective context length of 1 million tokens.
- How can developers deploy Beam on AWS?
- AWS describes deploying Beam through a SageMaker JumpStart real-time endpoint for a managed path, or through SageMaker HyperPod Inference with JumpStart, SGLang, or vLLM for more control over distributed GPU infrastructure.
- Where is Beam available at launch?
- AWS says Beam is available in SageMaker JumpStart in the US East (N. Virginia) and US East (Ohio) regions.
Sources
- Introducing Reflection Beam: an open-weight model for agentic coding on AWS | Artificial Intelligence — Amazon Web Services (AWS)