A new US lab called Reflection just shipped its first model, and it’s open. The hook: Ollama is still running it free for a limited time, so the window to try a 501B model without your own GPUs is now. Beam is a 501B-parameter mixture-of-experts model with 23B active per token, a 1M-token context window, and an Apache 2.0 license. It’s on Hugging Face now.
It’s aimed at coding and agent work. Quick hits from the Hugging Face post:
- 52 layers, 256 experts per MoE layer, 8 routed per token. Text in, text out.
- BF16, FP8 and NVFP4 checkpoints.
- Tool calling with its own grammar, and five reasoning levels from low to max.
- Day-0 support in transformers, vLLM and SGLang, plus its own Rust harness.
- Reflection is giving $10,000 in Hub credits to the best projects built on it.
You’re not running 501B on a laptop. But Ollama is a day-zero partner and has Beam on Ollama Cloud free for a limited time, with zero data retention, so you can try it from your terminal without downloading anything.
What I like is the license. Apache 2.0 at frontier scale means teams can actually own and fine-tune it. More of this, please.
Happy Hacking!
