The mixture of experts model features 501B total parameters and 23B active parameters with a 1M token context window under the Apache 2.0 license. On Nvidia Blackwell Ultra GPUs, Beam delivers over 70K tokens per second per GPU. Evaluations on AutomationBench-AA show a score of 48.5%, though the model recorded a 47% hallucination rate and a score of 3.0% on Terminal-Bench 4.0.
Wednesday, October 21, 2026
Story
Reflection's Beam Tops US Open Models on AI Cyber Index
AI5 days ago
Beam scored 30 on the Artificial Analysis Intelligence Index, placing it as the second-ranked U.S. open-weights model behind Meta's Muse Spark Open. The model's 34% score on the Cyber Index makes it the top-performing U.S. open-weights model tested in that category. In the Agent Arena, it is the highest-ranked U.S. model at number 12, while it ranks 22nd among open models in the Code Arena for WebDev as the sole U.S. entry in the top 25.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.