Reflection releases Beam, a 501B open-weight model, under Apache 2.0

Reflection has published the weights of Beam, its first open-weight model, on Hugging Face under the Apache 2.0 license. The first commit in the model’s repository, titled “Beam is available,” is dated October 16, 2026. Beam is a mixture-of-experts model with 501 billion parameters in total, about 23 billion of them active per token, and a 1 million token context window. Reflection announced the model on October 5 and said then that the weights would follow under Apache 2.0 “later this month.”

What is free

Apache 2.0 is a permissive open-source license that allows commercial use and changes to the model. The model card lists three versions of the instruction-tuned model, in BF16, FP8 and NVFP4 precision. Beam takes text in and gives text out, and the model card says it was “developed predominantly for English.”

The weights cost nothing, but the hardware to run them does, and the model card gives no guidance on how many GPUs you need. It links to quantized versions for tools such as llama.cpp, Ollama and LM Studio, and lists Baseten as a hosted inference provider.

What it costs as an API

If you would rather not host it, DigitalOcean’s Serverless Inference lists Beam at $0.40 per 1M input tokens, $0.04 for cache reads and $1.20 per 1M output tokens. Reflection’s October 5 post described early access to its own platform through a waitlist and gave no API price.

How good is it

Reflection says Beam is “competitive with larger open models like GLM 5.2 and approaching Qwen 3.8-Max on coding and agentic tasks.” On advanced reasoning tests, it says Beam scores about the same as GLM-5.2 “while using 3–4× less inference compute.” It also concedes that “frontier open models like Kimi K3 remain ahead on raw capability.” These are Reflection’s own figures, not independent tests.

To compare Beam’s hosted price with other open and closed models, see our AI model price comparison or run the numbers in the token cost calculator.