Reflection AI, a startup backed by Nvidia, has unveiled its first open-weight artificial intelligence model, named Beam, positioning it as a direct challenger to lower-cost Chinese models that have been gaining ground in coding and agentic tasks.
What Beam is designed to do
Reflection said Beam is competitive with Chinese startup Z.ai's GLM-5.2 and is closing in on Alibaba's Qwen3.8-Max on coding and agentic benchmarks, the tasks where AI systems write software and carry out multi-step work with limited human direction. The company is pitching Beam as an open, customisable alternative to closed models for developers building AI-driven software tooling.
How big the model is
The startup said Beam contains 501 billion total parameters, the variables that determine how an AI system processes information, but activates only 23 billion of them for each task. That selective activation is meant to make the model faster and cheaper to run by using only part of its network at a time. By comparison, Z.ai's GLM-5.2 has about 744 billion total parameters with 40 billion active parameters.
Why the timing matters
The launch lands in the middle of an intensifying contest between American and Chinese AI labs over open-weight models. US technology firms are trying to fend off competition from Chinese open models that are cheaper, more customisable and increasingly capable of generating code at a level close to leading systems from OpenAI and Anthropic.
Reflection was founded in 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou and focuses on tools that automate software development, one of the fastest-growing commercial uses of AI. Earlier this year, the startup signed a deal with SpaceX for additional computing capacity at the Elon Musk-led company's Colossus 2 data centre, underscoring how access to compute has become a central battleground for AI companies.
Sources:
Nvidia-backed Reflection unveils first AI model to take on Chinese open models — Reuters
What Beam is designed to do
Reflection said Beam is competitive with Chinese startup Z.ai's GLM-5.2 and is closing in on Alibaba's Qwen3.8-Max on coding and agentic benchmarks, the tasks where AI systems write software and carry out multi-step work with limited human direction. The company is pitching Beam as an open, customisable alternative to closed models for developers building AI-driven software tooling.
How big the model is
The startup said Beam contains 501 billion total parameters, the variables that determine how an AI system processes information, but activates only 23 billion of them for each task. That selective activation is meant to make the model faster and cheaper to run by using only part of its network at a time. By comparison, Z.ai's GLM-5.2 has about 744 billion total parameters with 40 billion active parameters.
Why the timing matters
The launch lands in the middle of an intensifying contest between American and Chinese AI labs over open-weight models. US technology firms are trying to fend off competition from Chinese open models that are cheaper, more customisable and increasingly capable of generating code at a level close to leading systems from OpenAI and Anthropic.
Reflection was founded in 2024 by former DeepMind researchers Misha Laskin and Ioannis Antonoglou and focuses on tools that automate software development, one of the fastest-growing commercial uses of AI. Earlier this year, the startup signed a deal with SpaceX for additional computing capacity at the Elon Musk-led company's Colossus 2 data centre, underscoring how access to compute has become a central battleground for AI companies.
Sources:
Nvidia-backed Reflection unveils first AI model to take on Chinese open models — Reuters
