Reflection releases Beam open weights
A 501B mixture-of-experts model, 23B active, under Apache 2.0, built for coding and agent work.
Official source (opens in a new tab)
Reflection's post on 16 October 2026 says the weights for Beam are on Hugging Face under an Apache 2.0 license. Reflection describes Beam as its first open-weight model, a sparse mixture-of-experts with 501 billion total parameters and 23 billion active, built for coding, reasoning, and long-horizon agent work. The weights ship in BF16, FP8, and NVFP4, with a speculative-decoding drafter. Reflection lists day-zero support in Transformers, vLLM, SGLang, Unsloth, Axolotl, TokenSpeed, and NVIDIA Dynamo. A public beta of Reflection's own API is free to start at platform.reflection.ai. Reflection also describes Mirror, an open-source terminal harness trained alongside the model.
Who it's for: teams that want to self-host or customize a large coding model, not a chat app you open in a browser.
Fits: Best for Vibe Coding and Best Agent Builder as a model you can run. It does not change the order of those lists.