Nvidia-backed Reflection AI launches open-weight model Beam
Open-weight AI models from China are putting pressure on US tech companies. They cost less, are easier to customize, and their coding skills now come close to top products from OpenAI and Anthropic.
Nvidia-backed startup Reflection AI has launched its first open-weight model, Beam. The company says Beam will compete with Chinese models like DeepSeek and Kimi in coding and agent tasks, matching Zhipu AI's GLM-5.2 and getting closer to Qwen3.8-Max.
Beam has 501 billion total parameters, but only 23 billion turn on for each task. This sparse design makes it faster and cheaper to run. For comparison, GLM-5.2 has about 744 billion total parameters, with 40 billion active.
Reflection AI was founded in 2024 by former DeepMind researchers. It builds tools to automate software development and has partnered with SpaceX for extra computing power.
Access ChatGPT instantly from any app on macOS with a hotkey — no switching tabs, no subscriptions.
Kimi is a powerful AI assistant for students and professionals that tackles research, coding, and creative tasks.
Delivers top-tier open-source foundation models and blazing-fast APIs to power any AI application at scale
Comments
Loading comments…