What makes Beam different from other coding models?

Beam is a sparse mixture-of-experts model with 501 billion total parameters but only 23 billion active parameters. It targets agentic coding and long-horizon tasks like maintaining code, using tools, and completing multi-step workflows.

Where can I use Beam?

At launch, Beam is available in Amazon SageMaker JumpStart in US East (N. Virginia) and US East (Ohio).

What license does Beam use?

Reflection releases Beam's weights through Hugging Face under the Apache 2.0 license. Beam 1.0 is available in BF16, FP8, and NVFP4 checkpoint variants.

Large Language ModelsAI Coding AssistantsOpen-Source AIAmazon AI Blog8m agoOct 17, 2026, 01:00 JST

Reflection AI launches Beam, a 501B open-weight coding model

Send on LINE

3 Key Points

What happened
Reflection AI introduced Beam, an open-weight model for agentic coding and long-horizon tasks, with 501 billion total parameters, 23 billion active parameters, and an effective context length of 1 million tokens.
Why it matters
A model with 501 billion total parameters but only 23 billion active parameters uses fewer resources per token, so developers and researchers gain more control over the runtime, deployment environment, and customization process.

WHO IT HITSDevelopers and researchers building software engineering and agentic systems can now deploy Beam on AWS with control over runtime, environment, and customization. Enterprises running AI on AWS gain a new open-weight option for code and multi-step tasks.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

Reflection AI's Beam enters a market where open-weight coding models are gaining traction, but its sparse mixture-of-experts design—501 billion total parameters with only 23 billion active per token—is a deliberate attempt to balance capability with serving cost. The 1 million token effective context length is aimed at long-horizon agentic tasks that need to track state across many steps.

For AWS customers, Beam's launch in SageMaker JumpStart means they can deploy it without managing infrastructure, or use HyperPod Inference for Kubernetes-native control. This dual-path approach lets teams start managed and move to custom runtimes like SGLang or vLLM as their needs grow.

The model's availability in US East (N. Virginia) and US East (Ohio) suggests an initial roll-out focused on major US regions. Reflection's decision to release weights under Apache 2.0 and support BF16, FP8, and NVFP4 formats may lower adoption barriers for organizations with different hardware profiles.

FAQ

What makes Beam different from other coding models?: Beam is a sparse mixture-of-experts model with 501 billion total parameters but only 23 billion active parameters. It targets agentic coding and long-horizon tasks like maintaining code, using tools, and completing multi-step workflows.
Where can I use Beam?: At launch, Beam is available in Amazon SageMaker JumpStart in US East (N. Virginia) and US East (Ohio).
What license does Beam use?: Reflection releases Beam's weights through Hugging Face under the Apache 2.0 license. Beam 1.0 is available in BF16, FP8, and NVFP4 checkpoint variants.

Amazon AI BlogRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.