Reflection AI Beam Open-Weight MoE Model
Reflection AI introduced Beam on 5 October 2026, describing it as its first open-weight model, a sparse Mixture-of-Experts (MoE) model built for coding, reasoning and agentic workloads.
Tool Overview
Features, steps and FAQ below
Features
- ✓ Reflection says Beam is a sparse Mixture-of-Experts (MoE) model with 501 billion total parameters and 23 billion active, built for coding, reasoning and agentic workloads
- ✓ Reflection says it pretrained the model on 23.8 trillion curated tokens from the web and proprietary licensed datasets, then ran high-compute reinforcement learning with over 100 million rollouts on 10.5K NVIDIA GB300 GPUs over four weeks with a 256K max context
- ✓ Reflection says Beam is competitive with larger open models such as GLM 5.2 and approaching Qwen 3.8-Max on coding and agentic tasks, while acknowledging Kimi K3 remains ahead on raw capability and Beam's advantage is inference-time efficiency
- ✓ Reflection says Beam matches GLM 5.2 on advanced reasoning benchmarks while using 3 to 4 times less inference compute, with the efficiency gap even larger against 2T+ parameter models
- ✓ Reflection says Beam offers a controllable reasoning effort parameter to trade off speed and performance; Beam is text-only and its weights will be released under an Apache 2.0 license this month
How to Use
- Visit the official Reflection Beam page (reflection.ai/blog/introducing-beam) to learn about the model's capabilities, benchmarks and training method
- Reflection says Beam is available to a select group of users for now; you can join the waitlist at platform.reflection.ai for early access
- Watch for the weights, technical report, model card and developer artifacts the company will release later this month, including the full stack for running, evaluating and fine-tuning
- Once open-sourced, deploy and evaluate Beam under an Apache 2.0 license within your own workflows
FAQ
What is Beam?
Reflection says Beam is its first open-weight model, introduced on 5 October 2026, a sparse Mixture-of-Experts model with 501 billion total parameters and 23 billion active.
How capable is Beam?
Reflection says it is competitive with GLM 5.2 and approaching Qwen 3.8-Max on coding and agentic tasks, and matches GLM 5.2 on advanced reasoning while using 3 to 4 times less inference compute.
Can I download Beam today?
Reflection says Beam is available to a select group of users with a waitlist, and that the weights, technical report and model card will be released this month under an Apache 2.0 license.
Is Beam open source?
Reflection says it will release the weights under an Apache 2.0 license in October 2026, along with documentation, a technical report and the full stack for running, evaluating and fine-tuning the model.
Does Beam support multimodal input?
Reflection says Beam is text-only but can work with other modalities when represented as text; demos include building a 3D game in p5.js and a live NYC subway dashboard from public data.