AI startup Reflection on Monday, Oct. 5, unveiled Beam, its first open-weight model: the company plans to publish the model's trained files, so anyone can download it and run it on their own computers. Reflection says it will release them this month under the Apache 2.0 license, a license that lets anyone use, change and share them free of charge, along with a technical report and a model card, the company's published overview of how a model was built and tested. Until then, Beam is still in final testing, and only a select group of early users can try it. Others can sign up for a waitlist.
Beam is large. It has 501 billion parameters, the numbers a model tunes as it learns. But it is a mixture-of-experts model, a design that switches on only part of the model for each piece of text, so it runs faster than its full size suggests: 23 billion parameters work at a time. Reflection says this efficiency is Beam's edge. It says Beam scores about as well as GLM-5.2, a larger model from the company Z.ai that anyone can download, on hard reasoning tests while using a third to a quarter of the computing power to produce its answers, though its own table shows GLM-5.2 slightly ahead on each of those tests.
Reflection built Beam mainly for writing and fixing software and for tasks it carries out on its own over many steps, such as searching the web and using software tools. It reads and writes text only. The company says it sharpened those skills through reinforcement learning, learning by trial and error with a reward for getting it right, running 10,500 Nvidia GB300 chips for four weeks on more than 100 million practice attempts.
In Reflection's own tables, Beam beats Inkling, the open model from Thinking Machines Lab, on the four coding tests where both have scores, though Inkling can work with more than text. The strongest open models stay ahead. On Terminal-Bench 2.1, a coding test, Reflection's table gives Beam 80.1, against 88.3 for Kimi K3 and 90.6 for DeepSeek V4.1 Flash, and Reflection says Kimi K3 remains ahead on raw ability.
Beam's scores come from Reflection, and no outside group has checked them yet. The computing-power comparison is also an estimate: Reflection worked it out from how much of each model is switched on at a time and how much text it writes, and says it is not a measure of actual running costs.
It is not yet clear what Beam will cost through Reflection's own online service or what computers it will take to run. Reflection says it will publish its safety test results in the technical report, and that Beam is the first model in a series.