Reflection has unveiled Beam, its first open-weight AI model, as U.S. technology companies seek to compete with lower-cost Chinese systems such as DeepSeek and Kimi. The model contains 501 billion total parameters but activates 23 billion for each task, a design intended to reduce the computing and cost required to run it. Reflection says Beam is competitive with Z.ai's GLM-5.2 and is closing in on Qwen3.8-Max in coding and agentic tasks. Existing company-reported testing showed scores of 80.9 on SWE-Bench Verified and 80.1 on Terminal Bench v2.1, with three to four times less inference compute than comparable systems. The full weights and technical documentation are due later in October 2026 under the Apache 2.0 license, enabling independent testing. Reflection was founded in 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou and develops tools to automate software development. Earlier this year, it signed a deal with SpaceX for additional computing capacity at the Elon Musk-led company's Colossus 2 data center.