Why this score US AI lab founded in 2024 by former Google DeepMind researchers, building open-weight models. Authoritative for its own actions (announcements, licences, specifications); an interested party on the capability of its own models, where its claims are recorded as statements and corroborated independently before use.
This month, we will release the weights under an Apache 2.0 license
Kept. Reflection released Beam's weights on Hugging Face under Apache 2.0, and Artificial Analysis lists them as released on 16 October with that licence — inside the month promised.
Beam advances the Western open-weight frontier and is competitive with larger open models like GLM 5.2 and approaching Qwen 3.8-Max on coding and agentic tasks. Where frontier open models like Kimi K3 remain ahead on raw capability, Beam's advantage is efficiency at inference time.
Mostly not borne out on the independent index. Artificial Analysis scores Beam 30 against GLM-5.2's 34, Qwen3.8 Max's 45 and Kimi K3's 44, and below Meta's Muse Spark Open (39) among US models. The concession that Kimi K3 leads holds; the index is broad, not coding-only.