Open-weight & owned stack · training-tuning
Mostik demonstrated a method for sharing internal states between AI models, linking GLM-5.2 and Qwen-3.5.
Mostik unveiled a method for AI models to share internal states, demonstrating it by linking Z.ai's GLM-5.2 with Alibaba's Qwen-3.5.
Read the original at forklog.comOpens the publisher's site in a new tabMore in Open-weight & owned stack
Abliteration.ai released an API-accessible GLM 5.2 derivative with safety refusals removed for offensive-security use.Sep 1MBZUAI’s Institute of Foundation Models launched six fully open K2 Horizon models, releasing weights, code, training data, and methodology.Sep 3Thinking Machines Lab released Inkling-Small, a 276-billion-parameter open-weight model with 12 billion active parameters.Sep 5Moonshot AI released Kimi K3, its reported 2.8-trillion-parameter open-weight model, with weights and technical infrastructure.Sep 2Alibaba released Qwen3.8-Max through its API, upgraded the 0902 snapshot, and described a forthcoming open-weights release.Sep 2