Reflection's Beam becomes the most capable open-weight model built outside China
Source: The Decoder (opens in a new tab) · Jonathan Kemper
Intel Summary
Reflection has released Beam, its first open-weight model based on a mixture-of-experts architecture. The model activates 23 billion of its 501 billion total parameters per token and aims to match GLM 5.2 on coding and reasoning while utilizing three to four times less compute.
Why It Matters
The release provides an open-weight foundation model engineered specifically for compute efficiency, offering builders and enterprises a non-Chinese open-weight alternative designed to reduce inference serving costs compared to larger dense or less efficient architectures.
Part of an ongoing development
SourceReflection AI launches Beam 501B open-source model
Reflection AI Inc. has launched Beam, an open-source large language model featuring 501 billion parameters. The release follows the startup reaching a $25 billion valuation and reportedly entering a $6.3 billion agreement with SpaceX to rent Nvidia GB300 NVL72 systems used for the initiative. Claims are as reported; this summary makes no determination about accuracy or significance.
More coverage of this development
Organizations & Entities
- China
- DeepSeek
Related Intelligence
- ReportSame development
Reflection AI debuts open-source Beam model with 501B parameters
Reflection AI Inc. has launched Beam, an open-source large language model featuring 501 billion parameters. The release follows the startup reaching a $25 billion valuation and reportedly entering a $6.3 billion agreement with SpaceX to rent Nvidia GB300 NVL72 systems used for the initiative.
SiliconANGLE - DevelopmentDevelopingAlso involving China
OpenAI attributes model reasoning extraction attempt to Moonshot AI
The company attributed a portion of this extraction activity to Chinese artificial intelligence startup Moonshot AI, according to reporting from CNBC. Claims are as reported; this summary makes no determination about accuracy or significance.
2 independent sources - DevelopmentNewAlso involving DeepSeek
DeepSeek releases TileLang open-source programming software for Huawei Ascend chips
DeepSeek and Huawei have released open-source programming tools for Huawei's Ascend AI hardware, centered around a programming language called TileLang. The tooling is designed to offer a simpler programming model than Nvidia's CUDA and optimize software performance on domestic Chinese AI chips. Claims are as reported; this summary makes no determination about accuracy or significance.
1 reporting source - DevelopmentNewAlso involving China
U.S. authorities arrested Greg Lui for smuggling Nvidia AI chips to China
U.S. authorities arrested Greg Lui, owner of San Gabriel Valley-based Earthmade Computer Inc, for allegedly conspiring to smuggle over $300 million worth of servers containing export-restricted Nvidia AI chips to China. Claims are as reported; this summary makes no determination about accuracy or significance.
1 reporting source