October 7, 2026
- Reflection released Beam, a 501B-parameter MoE activating 23B per token, claimed to match GLM 5.2 on coding and reasoning at three to four times less compute and billed as the most capable open-weight model built outside China (The Decoder). · AI & Model Security
in Pre-Auth File Read Hits Eight Atlassian Data Center Products