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

Reflection released Beam on October 16, its first open-weight model, built for coding, reasoning and long-horizon agent work. It is a sparse mixture-of-experts model with 501 billion total parameters and 23 billion active. The weights are on Hugging Face under the Apache 2.0 license, alongside a model card with benchmarks and a safety technical report.

Reflection is releasing BF16, FP8 and NVFP4 weights plus a speculative decoding drafter model, with day-zero support in Transformers, vLLM, SGLang, Unsloth, Axolotl and NVIDIA Dynamo.

It also open-sourced Mirror CLI, a terminal coding harness trained alongside Beam, with resumable conversations, parallel agents in separate worktrees, skills and MCP connections. Reflection says Beam has also been tested with OpenCode, Pi and Hermes.

Beam is available through Reflection's API (free to start in public beta) and on platforms including OpenRouter, Ollama, LM Studio, Together AI, Fireworks AI, Cloudflare and Vercel.