Reflection Open-Sources 501B-A23B MoE Model Beam and Launches on AWS
Understand the story
On October 6, 2026, Latent Space AINews reported that Reflection released Beam, an open-source model trained from scratch, using a MoE architecture with 501B total parameters and 23B activated parameters, released under the Apache 2.0 license, with full weights planned to be available within that month. On October 16, 2026, Hugging Face Blog (first-hand) reported that Reflection open-sourced its first model Beam on Hugging Face (reflectionai/Beam-1.0-501B-A23B), with parameter scale, MoE architecture, and Apache 2.0 license consistent with earlier reports, noting that the model supports 1M token context and focuses on coding and agent use cases. The same day, AWS Machine Learning Blog (first-hand) reported that the Beam weights model landed on AWS, offering deployment guides for SageMaker JumpStart and HyperPod, positioned for agentic coding and long-horizon tasks with an effective context length of 1 million tokens, weights released via Hugging Face under the Apache 2.0 license, with BF16, FP8, and NVFP4 checkpoints available. All reports are consistent on parameter scale, architecture, license, and context length; earlier reports described Beam as a text-only coding model, while later reports say it focuses on coding and agent use cases. Model weights are now available on Hugging Face and deployable on AWS.
AI overview of reporting · Updated 2 days ago
Developments
3 developments
- 10/16 23:23 · 1 reportsReflection AI Open-Sources Coding Model BeamAWS Machine Learning Blog:Introducing Reflection Beam: an open-weight model for agentic coding on AWS
- 10/16 08:00 · 1 reportsReflection Releases 501B Open-Source Model BeamHugging Face Blog:Welcome Beam: a 501B open model you can own, customize, and scale
- 10/6 14:28 · 1 reportsReflection releases open-source model Beam (501B-A23B)Latent Space:[AINews] Reflection Beam - 501B-A23B American Open Model
Reporting timeline
Explore different aspects of the story through its reporting.
- AWS Machine Learning BlogSelectedIntroducing Reflection Beam: an open-weight model for agentic coding on AWS
Reflection AI released its open-weight model Beam, built for agent coding and long-horizon tasks, with 501B total parameters, 23B activated parameters, and an effective context length of 1 million tokens. The weights are published on Hugging Face under the Apache 2.0 license, with BF16, FP8, and NVFP4 checkpoints available.
- Hugging Face BlogSelectedWelcome Beam: a 501B open model you can own, customize, and scale
Reflection has open-sourced its first model, Beam (reflectionai/Beam-1.0-501B-A23B), on Hugging Face: a 501B-parameter MoE model (23B activated per token) with 1M token context, released under the Apache 2.0 license and targeting coding and agent use cases.
- Latent Space[AINews] Reflection Beam - 501B-A23B American Open Model
Reflection released Beam, a text-only coding model with 501B total parameters and 23B activated, a MoE trained from scratch, with full Apache 2.0 weights to be released within the month.
Story heat trend
Current heat3·Comparable peak16(10/17 00:00)·Comparable change over 24 hours -70%
The trend compares the same continuously observed sources, which may cover fewer sources than the current heat score. Hover or select the chart to inspect each hour; use the left and right arrow keys to navigate.