Updated

Beam

Reflection released Beam’s open weights on October 16 under the Apache 2.0 license, following its earlier early-access announcement. The 501B-parameter sparse MoE model has 23B active parameters and is built for coding, reasoning, and agentic workloads; it can be self-hosted or accessed through Reflection and partners.

By No Code MBA · Facts checked

Provider
Reflection
Announcement date
Status at last check
Generally available
Model family
Beam
Capabilities
Text, Coding

What changed

Beam is Reflection's first open-weight model, a sparse mixture-of-experts with 501.3B total and 23.4B active parameters, 256 experts and a 1M-token context window. It is text in, text out and developed mainly for English, with a June 30, 2026 knowledge cutoff. Weights ship in BF16, FP8 and NVFP4 plus a speculative decoding drafter. Reflection reports 81.6% on SWE-bench Verified, 80.1% on Terminal Bench v2.1 and 90.6% on GPQA Diamond (BF16).

Access and availability

Download weights from Hugging Face (reflection/Beam-1.0-501B-A23B) under Apache 2.0, or call model Beam-1.0-501B-A23B through Reflection's OpenAI-compatible API at api.reflection.ai, which is free to start in public beta. It is also offered through partners including AWS, Google Cloud, Microsoft Foundry, Together AI, Fireworks, OpenRouter and Ollama.

Official access documentation

What you can build

Compare the released weights or a hosted endpoint on a bounded coding task from your own workflow, tracking successful completion, tool-use errors, latency, and review effort.

Limitations

Text only and developed predominantly for English. Some benchmark scores are lower on the FP8 and NVFP4 variants, and self-hosting hardware requirements were not stated.

Sources and release history