Reflection debuts Beam, a 501B open-weight AI model

Reflection AI said Monday in a blog post that Beam, its first open-weight model, is a sparse mixture-of-experts system with 501 billion total parameters, 23 billion of them active, built for coding, reasoning and agentic work.

Who can use it: Reflection is making an early version of Beam available to a select group of users, with a waitlist at platform.reflection.ai. The model is still undergoing final red-teaming and evaluations, the company said. Beam is text-only.

The open part: Reflection said it will release the weights this month under an Apache 2.0 license, along with a technical report, model card, documentation and the full stack for running, evaluating and fine-tuning the model.

How it compares: Reflection says Beam is competitive with larger open models like GLM 5.2 and approaching Qwen 3.8-Max on coding and agentic tasks, while Kimi K3 remains ahead on raw capability. On advanced reasoning benchmarks, the company says Beam scores comparably to GLM 5.2 while using “3–4× less inference compute,” which it calls an approximate comparison rather than measured inference cost. Reflection reports Beam scored 80.9 on SWE-bench Verified and 90.5 on GPQA Diamond.

How it was built: Reflection said it pretrained Beam on 23.8 trillion tokens, then ran reinforcement learning on 10,500 Nvidia GB300 GPUs for four weeks, generating more than 100 million rollouts.

What’s next: Reflection says the weights, technical report and model card arrive later this month, and that Beam is the first model in a series.

— Rowan Keel