Z.ai launched GLM-5, a 744B-parameter mixture-of-experts model with about 40B active parameters, released with open weights under the MIT license. It targets complex systems engineering and long-horizon agent tasks and uses DeepSeek Sparse Attention to reduce long-context serving costs. For practitioners, it offered a permissively licensed frontier-scale option for self-hosted coding and agent workloads.
Alibaba unveiled Qwen3.8-Max, a sparse mixture-of-experts model with 2.4 trillion total and about 95B active parameters aimed at coding and 'cowork' agent tasks, and committed to…
Alibaba Qwen
Moonshot AI published weights for Kimi K3, a 2.8-trillion-parameter mixture-of-experts model with about 104B active parameters, a 1M-token context window and native text, image…
Moonshot AI
Z.ai released GLM-5.2, a roughly 753B-parameter successor to GLM-5.1 with a 1M-token context window and multiple thinking-effort levels, under an MIT license without regional…
Z.ai
DeepSeek published open weights for DeepSeek-V4-Pro (1.6T total, 49B active parameters) and V4-Flash (284B total, 13B active), both with 1M-token context and thinking and…
DeepSeek
