Reflection's Beam: A US-Built Open-Weight Coding Model You Can Try Free on Ollama

Reflection AI, a New York startup founded by former DeepMind researchers, has released its first open-weight model, called Beam. Ollama, the popular tool for running AI models, announced on October 16 that Beam is available through Ollama Cloud and free “for a limited time”, with no end date given. The model is aimed at software engineering, reasoning and long-running agent tasks.

Here are the specs from Ollama’s page. Beam has about 501 billion parameters in total, but it is a “mixture-of-experts” design, so only about 23 billion are active for any one token (a token is a small chunk of text). Its context window, roughly how much text it can keep in mind at once, is 1 million tokens. The licence is Apache 2.0, which allows you to inspect, change and build on the model. To try it, update Ollama and run ollama run beam:cloud; the page also shows commands to use it from tools such as Claude Code, Codex and OpenCode.

On quality, Reflection says Beam is comparable to Z.ai’s GLM-5.2 and approaches Alibaba’s Qwen 3.8-Max, while both still trail the best closed models from OpenAI and Anthropic. CEO Misha Laskin claims it needs three to four times less compute to reason through a problem than comparable open models. Those numbers come from Reflection, and there is no independent verification yet. The coverage also does not give benchmark scores or pricing after the free period.

What’s behind it: “open-weight” means the finished model files are published, so others can run and adapt them, even if the training data and recipe stay private. Until recently, many of the strongest open-weight models came from Chinese labs. Reflection pitches Beam as a US-built alternative for developers who want that control without relying on those models, and Nvidia is among its backers. The mixture-of-experts trick is why a 501-billion-parameter model can be affordable: like a hospital where each patient sees only the relevant specialists, not every doctor.

What this means for you: if you are new to this, the practical news is that you can test a large open model in a few minutes without owning a powerful computer, because it runs in Ollama’s cloud. Note that this means your prompts go to the cloud, so it is not “local” in the privacy sense. If you are more experienced, the Apache 2.0 licence makes Beam interesting for products where licence terms matter. Try it on a real task of your own before believing any benchmark, and do it soon, since the free period can end without warning.

Sources

• Ollama: Reflection’s Beam is freely available on Ollama for a limited time
• DeepLearning.AI, The Batch: Reflection releases low-cost LLM with open weights