Beam FAQ
What is Beam?
Beam is Reflection AI’s first open-weight model: a sparse Mixture-of-Experts model with 501B total and 23B active parameters, aimed at coding, reasoning and agentic workloads.
When will the Beam weights be released?
Reflection AI says the weights, technical report and model card will be published Late October 2026. The model is currently in final red-teaming and evaluation.
What license does Beam use?
Apache 2.0, which allows commercial use, modification and redistribution.
How long is Beam’s context window?
Beam was extended to 1M tokens during midtraining.
Can I run Beam on a single GPU?
Not at full precision. The FP8 weights take about 500 GB. Aggressive GGUF quantizations may run on large-memory Macs or multi-GPU workstations; use the VRAM calculator to check your setup.
Why is Beam fast if it has 501B parameters?
Only 23B parameters are active for each token, so compute per token is closer to a 23B dense model. Reflection AI reports scores comparable to GLM-5.2 at 3–4× less inference compute.
Is this the official Beam website?
No. BeamModel.com is an independent community resource. Official information is published by Reflection AI at reflection.ai.