Reflection releases Beam, a 501B open-weight model
The startup's first frontier-scale model ships with downloadable weights and a mixture-of-experts design aimed at reasoning.
Reflection has introduced Beam, a 501-billion-parameter language model released with downloadable weights. According to the company's announcement, the model is positioned as a frontier-scale system built around text generation and reasoning.
Beam uses a mixture-of-experts (MoE) architecture, a design that activates only a subset of its parameters for any given token. That approach lets a model carry a very large total parameter count — here, 501B — while keeping the compute cost of each inference step more manageable than a dense model of equivalent size.
Why it matters
Open-weight releases at this scale remain relatively rare, and most frontier-class reasoning models are kept behind APIs. Making the weights available changes who can study, fine-tune, and deploy the model.
- Open weights: the model can be downloaded and run outside a vendor's API.
- MoE efficiency: a large total parameter count with sparser active compute.
- Reasoning focus: positioned for text and reasoning workloads rather than a single narrow task.
The release is licensed under terms the company describes separately, so teams considering commercial use should review the specifics before building on it. For a group's first frontier-scale entry, Beam is a notable signal that open-weight development continues to push into larger, reasoning-oriented territory.
Sources
More from OpenAI
All OpenAI releases →Aleph Alpha launches Kolibri, a sovereign open model
The German AI firm positions its new open-weight language model as a European alternative built for data control and compliance.
H Company releases Holo4 for computer-use agents
The new vision-language model is built to drive generalist agents that operate real software interfaces.
Thinking Machines ships Inkling, its first open model
The Mira Murati-founded lab makes its debut with an open-weights, reasoning-focused language model.
More in Text / LLM
All Text / LLM →
Microsoft's Aion 1.0 Instruct is a compact 2.3B chat model
A small, multilingual instruction-tuned model aimed at chat and summarization lands on Hugging Face.
SenseTime Releases Looped 9B Multimodal Model
The company's new SenseNova-Loop-9B pairs vision-language capabilities with a recurrent looped-transformer design.

Tencent releases Hunyuan Hy4-flash multimodal MoE
The lightweight, mixture-of-experts model adds vision and text instruction-following to Tencent's open Hunyuan lineup.
0 comments
No comments yet. Be the first to weigh in.