According to GoogleDeepMind, the 740M EmbeddingGemma 2 beats larger specialists, enables multimodal search, and ships under Apache 2.0.

According to @StanfordAILab, SWE-chat v2 publishes 230K real-world coding prompts on HuggingFace, enabling training and evaluation of AI code agents.

According to TheRundownAI, Z.ai launched GLM-5.3-Flash on OpenRouter, topping weekly use with $0.15/M in, $0.50/M out; weights released on HuggingFace.

According to @openclaw, OpenClaw now runs as a fully local tool-calling agent on HuggingFace apps using GGUF or MLX models with no cloud or keys.

According to @openclaw, scans of 67,453 ClawHub skills show 0.31% malicious and under 8.5% agreement across tools, signaling risk taxonomy gaps.