#huggingface

DeepSeek releases V4.1‑Flash‑1017 model on Hugging Face, boosting inference speed with FP8 precision and hinting at up to 700B parameters.

Explore Prism ML’s new Ternary Bonsai 2 LLMs—2B, 4B, 9B models that run fast in browsers with WebGPU. Download now on Hugging Face.

Black Forest Labs’ Flux.3 Image Klein delivers 5.6B‑parameter text‑to‑image generation and multi‑reference editing—open‑weights, Qwen3‑VL‑4B‑Instruct encoder, ready for Hugging Face or ComfyUI deployment.

Explore NVIDIA’s new Nemotron‑3.5 Super VL 120B AI model: hybrid Mamba‑2/Transformer architecture, 1‑M token context, and efficient Bolt‑74B variant for code generation and RAG tasks.

Explore why dual NVIDIA DGX Spark clusters hit memory and bandwidth limits for large LLMs—real benchmarks, experimental tricks, and best‑practice guidance.

LiquidAI unveils the 24‑B LFM2.5‑24B‑A2B—an on‑device, 2.45 B‑parameter MoE model that rivals larger rivals while running faster, enabling multilingual private agents.

Anthropic’s new research on self‑replicating AI sparks debate over open‑weight model security, control, and potential regulatory backlash—uncover why it matters.

Discover vLLM’s Decision 3.0 multimodal models—up to 27B parameters—delivering precise choice, yes/no, or score answers for text, images, and video without text generation. Explore the new family, top Jev benchmark wins, and Hugging Face availabilit…

Upcycling dense LLMs to MoE on an RTX 4060 cuts MLP compute 40‑44% but drops accuracy—proof of concept for low‑resource, sparse model conversion.

Explore Tencent’s Youtu‑Parsing‑Omni: a 5B‑parameter omni‑modal model that turns text, images, audio, and video into structured JSON with top benchmark performance.