Reflection Beam
Open models you can inspect, fine-tune, and deploy on your own terms.
Open-weight 501B-parameter MoE language model for reasoning, coding, and agentic tasks with a 1M-token context window.
Reflection homepage — Introducing Beam, the 501B open-weight model
1 / 2Overview
What is Reflection Beam?
Open models you can inspect, fine-tune, and deploy on your own terms.
Reflection Beam is the first frontier open-weight model from Reflection AI, announced on October 5, 2026. It is a sparse mixture-of-experts language model with 501 billion total parameters and 23 billion active parameters per token, pretrained on 23.8 trillion tokens with a 1 million token context window.
Reflection trained Beam with a particular focus on coding, reasoning, and agentic workloads, using high-compute reinforcement learning at large scale. The company reports that Beam matches Z.ai's GLM-5.2 on advanced reasoning benchmarks while using 3-4x less inference compute, and that it advances the Western open-weight frontier on coding and agentic tasks. These are company-reported figures that have not been independently verified, since the weights are not yet public.
As of the announcement, Beam is undergoing final red-teaming and evaluations. Reflection says it will release the open weights under an Apache 2.0 license later this month, together with a technical report, model card, and developer artifacts, and offers early access through a waitlist on its website.
Platforms and languages
Reflection Beam Availability
Platforms
Languages
Capabilities
Reflection Beam Key Features
501B sparse mixture-of-experts
501 billion total parameters with 23 billion active per token; text-only model built for reasoning, coding, and agentic workloads.
1M-token context window
Handles very long documents and extended agent trajectories within a single context window.
High-compute reinforcement learning
Trained with large-scale RL — over 100 million rollouts on 10.5K NVIDIA GB300 GPUs over 4 weeks, per the company's announcement.
Open weights (planned)
Company plans to release weights under an Apache 2.0 license later this month, alongside a technical report and model card.
Built for self-hosting and fine-tuning
Designed so enterprises and governments can inspect, fine-tune on private data, and deploy in their own environments.
Best for
Who uses Reflection Beam?
Developers building coding agents
Evaluate an efficient open-weight model for code generation and agentic workflows
Enterprises and public-sector teams
Deploy a US-origin open-weight model fine-tuned on private data in their own infrastructure
AI researchers
Inspect and benchmark a frontier open-weight model once the weights are released
Common questions
Reflection Beam FAQs
Is Reflection Beam available to use right now?
Not yet. Beam was announced on October 5, 2026 and is in final red-teaming and evaluations. The company plans to release the open weights later this month and offers early access through a waitlist on reflection.ai.
How big is Beam and what can it do?
Beam is a text-only sparse mixture-of-experts model with 501 billion total parameters and 23 billion active per token. It has a 1M-token context window and was built for reasoning, coding, and agentic workloads.
What license will Beam use?
Reflection says it plans to release Beam's weights under an Apache 2.0 license, along with a technical report and model card. The license has not been published yet, so check the actual license file before building a product on it.
How does Beam compare to DeepSeek?
Reflection positions Beam against Chinese open models like DeepSeek. The company reports Beam matches Z.ai's GLM-5.2 on advanced reasoning benchmarks at 3-4x less inference compute, but these figures are company-reported and unverified — and DeepSeek's V4 weights are downloadable today while Beam's are not.
Who is Beam for?
Developers building coding agents, and enterprises or public-sector teams that want a US-origin open-weight model they can fine-tune on private data and deploy in their own environments.
Reviews
See what the community thinks and share your experience.
—
Based on 0 ratings
Rating distribution
Community reviews
No reviews
No written reviews yet.
