Reflection AI debuts Beam, its first open-weight model
Reflection AI, a 2024 startup founded by two former Google DeepMind researchers and backed by Nvidia, unveiled Beam, a text-only mixture-of-experts model with 501 billion total parameters and 23 billion active, pretrained on 23.8 trillion tokens with a 1 million token context window and aimed at reasoning, coding and agentic work. The company says Beam reaches performance close to Z.ai's GLM-5.2 using roughly one-third to one-quarter of the inference compute, and approaches Qwen3.8-Max, though its own published scores still trail newer Chinese open models such as GLM-5.3, Kimi K3 and DeepSeek V4.1 Flash. Reflection says the weights, a technical report and developer tooling are due this month.
It marks the first US startup attempt at a frontier open-weight model competitive with leading Chinese releases, though the efficiency claims are self-reported and unverified.