Release2026-10-16
Reflection released Beam 1.0, an open sparse mixture-of-experts model. It has 501B total parameters and 23B active parameters. It supports a 1M-token context window. It is available in BF16, FP8 and NVFP4 formats. It is designed for token-efficient reasoning, coding and agentic tool use. It was trained on web, code, technical and scientific data. It passed internal and third-party safety evaluations. It is licensed under Apache 2.0. It allows commercial and research use, including fine-tuning and distillation. Its knowledge cutoff is 30 June 2026.
What it unlocks
Running a 501B-parameter model with 23B active at a time for coding, reasoning and agent tasks.
- 501.3B total parameters
- 23.4B active parameters
- 1M token context window
- 2,930 downloads last month
What you need to act on it
- Apache 2.0 license
- commercial or research use allowed
Sources