Open-Weight Models Intensify Frontier Coding Competition

Mistral Large 4, Reflection AI Beam, and Tencent's reported 770B Hy4 expand the open-weight challenge, with long context, sparse MoE designs, local deployment, vulnerability patching, and autonomous workflows as key signals. Hy4 reportedly narrowly beats GLM-5.3 in Tencent-linked evaluations, but independent validation, inference economics, licensing, checkpoint availability, and production results remain unresolved.