Model updates

Official news, first rumor reports and substantive progress. Each entry keeps the information published at that time.

• Hugging Face Transformers 5.20.1 adds Reflection AI Beam support and fixes older-torch loading

  Transformers 5.20.1 adds native support for Reflection AI's Beam, the 501B-total / 23B-active open-weight MoE aimed at coding and agentic use, with model docs under the Beam page. The patch also restores from_pretrained on torch 2.6/2.7 by requiring torch>=2.8 only for distributed checkpointing, and fixes HunYuan-VL batched image assignment, DeepGEMM Mega MoE under expert parallelism, auto-mapping bugs and DtensorShardOperation on older torch.

  OfficialOfficial updateOct 16Source: Hugging Face TransformersPart of Beam
• Reflection AI unveils Beam, a 501B-parameter open-weight MoE model, weights due later this month

  Reflection AI introduced Beam, its first open-weight model: a sparse mixture-of-experts model with 501 billion total parameters, 23 billion active, built for coding, reasoning, and agentic workloads. It was pretrained on 23.8 trillion tokens, and the high-compute RL run generated over 100 million rollouts on 10.5K NVIDIA GB300 GPUs over four weeks. Reflection says Beam is competitive with larger open models like GLM 5.2 and approaching Qwen 3.8-Max on coding and agentic tasks, with its advantage being efficiency at inference time. The model is undergoing final red-teaming, and the weights, technical report and model card are due later this month; early access is open by sign-up.

  OfficialOfficial updateOct 5Source: Reflection AIPart of Beam
• Ollama says more US models are coming, reacting to Reflection AI's new open model

  The Ollama account posted that it is excited about the new open model from Reflection AI and that more US models are coming to Ollama. Reflection unveiled Beam, a 501B-parameter open-weight MoE model, on October 5, with weights due later this month.

  UnconfirmedPublished newsOct 5Source: X @ollamaPart of Beam

RSS for this view →