DevelopingOpen source
Reflection AI releases open-weight Beam, 501B parameters under Apache 2.0
Reflection AI released Beam on Oct. 16, a mixture-of-experts model with 501B total and 23B active parameters, with weights on Hugging Face in BF16, FP8 and NVFP4 under Apache 2.0. It also released a speculative-decoding drafter, a safety technical report and the Mirror CLI agent harness. Self-distillation recipes are promised in coming days.
- HEAT
- OUTLETS
- 1
- NEW IN 6H
- +1
- FIRST SEEN
- LAST SIGNAL
Primary source
Reflection AIreflection.ai/blog/beam-is-availableCoverage · 2 articles, oldest first
Friday, October 16, 2026
Reflection AIPrimary
Reflection AI releases open-weight Beam, 501B parameters under Apache 2.0