Reflection AI releases Beam, a 501B open-weight model, under Apache 2.0
Announced · Published · By Contractor AI Shop
Reflection released weights for Beam, its first open model, on Hugging Face under the Apache 2.0 license. It is a mixture-of-experts model with 501B total and 23B active parameters, a 1M-token context, and BF16, FP8 and NVFP4 checkpoints, built for coding and agents.
Key facts
- Reflection says Beam's weights are on Hugging Face under Apache 2.0, with a model card and a safety technical report. Reflection
- VentureBeat reports Reflection pitches Beam on finishing tasks in fewer tokens, which it says lowers cost per task. VentureBeat
- Reflection told VentureBeat the 8-bit and 4-bit versions fit on a single server of eight NVIDIA GPUs; full precision needs tens of GPUs. VentureBeat
- SiliconANGLE reports Reflection also open-sourced post-training recipes alongside the weights. SiliconANGLE
Our take: what it means for contractors
A permissive U.S. open model with a cleaner license than Meta's, but it needs a GPU server, not a desk machine. For contractors it is something you'd use through a hosted provider or a vendor's private deployment, not on a Mac Studio.
What we don't know yet
- Reflection's own hosted pricing (free during public beta)
- Independent cost-per-task results
- Whether community quantizations make it practical on Mac Studio clusters
Sources
- Reflection · Beam is now available · 2026-10-16 · Primary
- VentureBeat · Reflection's Beam aims to undercut Chinese open models with fewer tokens, not cheaper tokens · 2026-10-16 · Secondary
- SiliconANGLE · Reflection AI open-sources Beam's weights, post-training recipes · 2026-10-16 · Secondary