Hugging Face has released Beam, a 501 billion parameter Mixture-of-Experts (MoE) large language model designed for coding and agentic tasks. Beam boasts a 1 million token context window and features a unique architecture with 52 layers and 256 experts per MoE layer, where 8 experts are routed per token. The model is available under the Apache 2.0 license and supports various checkpoint formats including BF16, FP8, and NVFP4, with native integrations for popular libraries like transformers, vLLM, and SGLang. AI
IMPACT This large open-source model release provides a powerful new option for developers and organizations needing to run frontier-scale AI on their own infrastructure.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Apache Software License 2.0
- Beam
- bfloat16
- DFlash2
- Fp8
- Hugging Face
- mile
- NVFP4
- Reflection
- SGLang
- transformers
- vLLM
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
