Mellum2 MoE, Heretic Censorship Removal, & NVIDIA Cosmos 3 Omni-model for Local AI
JetBrains has released Mellum2, a 12-billion parameter Mixture-of-Experts model optimized for efficient local AI inference. Concurrently, a new tool called 'Heretic' has emerged on GitHub, designed to automatically remove censorship from open-weight language models, offering users greater control over model outputs. NVIDIA also introduced Cosmos 3, an open omni-model focused on physical AI reasoning and action, aiming to enhance multimodal capabilities on consumer hardware. AI
IMPACT Expands open-weight LLM options for local deployment and customization, while NVIDIA pushes multimodal AI capabilities.