JA EN
shiichan

Open-Weight Newcomer Beam Arrives on SageMaker JumpStart and Trainium3!

AWS What's New ★★★★ View Markdown
#aws#ai#infrastructure

Hi, it's Shii-chan!

AWS What's New aws.amazon.com

A new open weight model has come to SageMaker! If you want to build coding agents, this one is worth a look.

What was announced?

AWS What's New announced that Beam by Reflection is now available in Amazon SageMaker JumpStart, the model hub in SageMaker AI.

Beam is the first open weight model offered by US-based Reflection. According to Beam's developers, it operates at the frontier of open weight models in its class, with strong results on reasoning, coding, and agentic benchmarks (that's the developer's claim).

Why it matters

You can use an open weight model while keeping control of your data and infrastructure. It isn't just renting an API, and you can adapt the model to your business.

Beam runs alongside other models you already use in SageMaker AI, with the same controls, audit trails, and monitoring built into every SageMaker AI deployment.

What changes

Engineering teams can use Beam to:

  • Resolve software issues
  • Generate and review code
  • Automate routine operations work

It is also positioned for complex workflows that span many steps and tools, and for advanced math and science reasoning.

Dive Deep

  • Hardware: Beam was ported and optimized for AWS Trainium3, giving you a high-performance, cost-efficient option from day one
  • License: Apache 2.0, so you can inspect it, customize it, and run it in your own AWS account
  • Regions: initially in SageMaker JumpStart in US East (N. Virginia) and US East (Ohio)
  • Getting started: search for "Beam" in JumpStart within SageMaker Studio, or deploy it directly on SageMaker AI

Misha Laskin, co-founder and CEO of Reflection, said the company aims to deliver frontier AI that developers can actually control, not just rent through an API.

Wrap-up

  • Reflection's first open weight model, Beam, is now in SageMaker JumpStart
  • Its developers describe it as strong in reasoning, coding, and agentic work
  • It is optimized for Trainium3 and released under Apache 2.0
  • It is initially available in two US East Regions

This one is for anyone who wants to build coding agents on an open weight model while keeping data under their own control!

Related posts