Article · AI translation
IT Home reported on October 6 that Nvidia-backed startup Reflection AI unveiled its first open-weight large model, Beam, on Monday local time. The startup hopes to compete with Chinese AI models such as DeepSeek and Kimi in coding and agent tasks.
Reflection AI said Beam's performance is comparable to GLM‑5.2 from Chinese AI startup Zhipu AI (Z.ai); in coding and agent tasks, the model's performance is gradually approaching that of Qwen3.8‑Max.
The startup said Beam has 501 billion parameters in total; parameters are the variables that determine how an AI system processes information. However, the model uses a sparse activation architecture and activates only 23 billion parameters for each task. By using only part of the network at a time, it achieves faster inference while reducing operating costs.
By comparison, Zhipu AI's GLM‑5.2 has about 744 billion parameters in total, with 40 billion active parameters.
The model's release comes as U.S. technology companies seek to fend off competitive pressure from Chinese open-weight models. These models cost less, offer greater customizability, and have code generation capabilities that have nearly caught up with leading models from OpenAI and Anthropic.
According to IT Home, Reflection AI was founded in 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou. The company primarily develops tools for automating software development, which is also a rapidly growing category of AI application.
Earlier this year, Reflection AI reached an agreement with SpaceX to access additional computing resources at Colossus 2, a SpaceX data center operated by Elon Musk's company.