Reflection AI has unveiled Beam, its first open-weight model. Built on a Sparse Mixture-of-Experts architecture, the model has a total of 501 billion parameters, with 23 billion active per token. Focused on coding, reasoning, and agent-based tasks, Beam was pre-trained on 23.8 trillion tokens and offers a context window of 1 million tokens. During the reinforcement learning phase, 10 thousand 500 Nvidia GB300 GPUs were used for four weeks, more than 100 million rollouts were generated, and approximately 1.3 billion sandboxes were run.
According to the company’s own tests, Beam achieves performance similar to GLM-5.2 on challenging reasoning benchmarks with 3 to 4 times less inference compute; the results have not yet been independently verified. The model scored 77.2 on SWE Bench Pro v2-Hard, 80.1 on Terminal Bench v2.1, and 80.9 on SWE Bench Verified. The company acknowledges that larger open models such as Kimi K3 are ahead in raw performance.
Beam, which is text-only, can conduct research using tool access and web connectivity. Reflection AI, founded in 2024 by two former Google DeepMind researchers, has raised approximately $4.7 billion in investment; its valuation in the latest round was $25 billion. The model weights and developer tools will be released in October.
Background
Beam is not a new name in the FikirPilot archive: we have published 3 stories featuring this name in the last 90 days; the most recent is dated October 6, 2026.