Nvidia announced Cosmos 3 via X, a unified physical AI model using action-as-token.

RTX Pro 6000 Blackwell price raised to $13,250, a 55% increase

Nvidia qualified HBM4 for Vera Rubin from Samsung, SK Hynix, Micron

NVFP4 achieves 1.8x training speedup over FP8 on GPT-3 175B model with no accuracy loss claimed for models up to 70B.

NVIDIA announced NVFP4 4-bit precision format for Blackwell GPUs integrated into MaxText.

Qualified HBM4 from Samsung, SK Hynix, and Micron for Vera Rubin

Released Unsloth Studio, an open-source web app for local LLM training with reduced VRAM usage

Released method to run Claude Code locally using open-source LLMs via environment variable workaround.

Publishes guide and tools for connecting VS Code to Google Colab for LLM fine-tuning.