Nvidia-backed startup Reflection AI has launched Beam, its first open-weight artificial intelligence model, as the U.S. company looks to challenge lower-cost Chinese AI models in coding and agentic tasks.
Reflection AI said Monday that Beam delivers performance competitive with Chinese AI developer Z.ai’s GLM-5.2 and is narrowing the gap with Qwen3.8-Max. The model is designed primarily for coding and AI agent applications, areas where open-weight systems from China have increasingly gained traction.
Beam contains 501 billion total parameters, the variables that help determine how an AI model processes and generates information. However, only 23 billion parameters are activated for each task. This approach allows Beam to use only a portion of its network at a time, potentially reducing computing requirements while improving speed and lowering operating costs.
By comparison, Z.ai’s GLM-5.2 has approximately 744 billion total parameters and activates about 40 billion parameters during processing.
The Beam launch comes as U.S. artificial intelligence companies face growing competition from Chinese open-weight models, including systems developed by DeepSeek, Kimi, Qwen and Z.ai. These models have attracted attention because they can be cheaper to operate, offer greater customization and generate computer code at performance levels approaching leading proprietary models from OpenAI and Anthropic.
Reflection AI was founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou. The startup focuses on developing AI tools capable of automating software development, which has emerged as one of the fastest-growing commercial applications for generative AI.
Reflection has also been expanding its computing infrastructure to support development of increasingly powerful AI models. Earlier this year, the company reached an agreement with SpaceX for additional computing capacity at the Elon Musk-led company’s Colossus 2 data center.
With Beam, Reflection AI is positioning itself in the intensifying global race to develop powerful, cost-efficient open-weight AI models capable of handling advanced coding and autonomous agent tasks.


OKXICE Files for 24/7 Tokenized US Stock Trading
Coinbase, Circle Stocks Slide as Bank Group Sues OCC Over Crypto Charters
Nvidia China Chip Sales Report Sends Chinese Semiconductor Stocks Lower
Trump, Tech Leaders Sign ‘Morally Binding’ AI Safety Accord
Meta and Microsoft Cut Internal Claude AI Use
Fairshake Backs 32 Pro-Crypto House Candidates
DeepSeek, Huawei Partner on Ascend AI Chip Programming Tools
Polymarket Traders Bet on Trump AI Rename to ‘Super Intelligence’
BlackRock Sees AI Agents Fueling New Crypto Demand
Samsung Invests $1 Billion in KKR’s Helix AI Infrastructure
CSL Partners With AWS to Bring AI Into Drug Research
Schneider Electric to Buy PTC in $22.6 Billion Deal
Micron Forecast Tops Estimates as AI Memory Demand Stays Strong
Revolut Challenges Europe’s Banking Giants as Global Growth Accelerates
TD Cowen Starts SpaceX at Buy, Sees AI Compute Driving Growth
Zcash Mining Profits Surge as Liquidity Risks Grow 



