AI BriefingHuggingFace Blog news
The latest AI and developer news from HuggingFace Blog, with short summaries.
Visit website →Diff-Transformer V2 Released
Differential Transformer V2 has been released, improving inference efficiency by expanding query heads while keeping the number of KV heads unchanged.
Real-Time Interactive Video Model Waypoint-1
Overworld has released Waypoint-1, a video diffusion model that supports real-time control via text and keyboard/mouse, along with the inference library WorldEngine.
Open Reasoning Standard for Agents Unveiled
Open Responses, a new open reasoning standard optimized for agentic workflows, has been unveiled.
NVIDIA Unveils Cosmos Reason 2 for Physical AI
NVIDIA has released Cosmos Reason 2, an open VLM with enhanced reasoning capabilities for physical AI.
Falcon-H1-Arabic Model Released
Falcon-H1-Arabic, an Arabic-specialized model with a hybrid architecture combining Mamba and Transformer, has been released.
NVIDIA Releases Guide for Building Robot Agents
NVIDIA has revealed how to build AI agents that operate in physical environments using DGX Spark and Reachy Mini.
ServiceNow-AI Unveils AprielGuard, a Safety Model for Agents
AprielGuard, an 8B-parameter guardrail model designed for the security and safety of agentic workflows, has been released.
Transformers v5 Overhauls Tokenizer Architecture
Transformers v5 significantly strengthens modularity and customization convenience by separating tokenizer design from trained vocabulary.
NVIDIA Unveils Open Evaluation Standard
Alongside the launch of Nemotron 3 Nano, NVIDIA unveiled an open evaluation standard built on NeMo Evaluator for transparent verification.
High-Performance Open-Source AI Agent 'CUGA' Released
CUGA, an open-source AI agent framework that performs complex web and API tasks, has been released.
llama.cpp Introduces Dynamic Model Management Feature
A router mode has been added to the llama.cpp server that allows multiple models to be dynamically loaded and switched without restarting.
Automating AI Model Training and Deployment with Codex
Hugging Face has released HF-skills, which helps coding agents perform model training, evaluation, and deployment.
Hugging Face Releases Swift Client
Hugging Face has launched swift-huggingface, a dedicated client for Swift developers that supports model downloads and the Hub API.
Hugging Face Unveils 'Hugging Face Skills' That Let Claude Train LLMs Directly
Hugging Face has launched 'Hugging Face Skills,' which allows AI agents to directly perform LLM fine-tuning.
Intel Unveils Lightweight Math Agent DeepMath
Based on Qwen3-4B, DeepMath uses Python code to improve math reasoning accuracy while cutting output length by 66%.
Hugging Face Unveils Transformers v5
Hugging Face has unveiled Transformers v5, which strengthens simplicity and efficiency in model definitions.
LLM Inference Optimization: How Continuous Batching Works
This piece explains in detail, from the fundamentals, how continuous batching works to maximize LLM inference efficiency.
Black Forest Labs Unveils FLUX.2
Black Forest Labs' new image generation model FLUX.2 has been released and is available in the Diffusers library.
Tavily Unveils Strategy for Building High-Performance Research Agents
It explains agent harness and context engineering strategies for maximizing the performance of research agents.
OVHcloud Joins Hugging Face as Inference Partner
OVHcloud has joined as an Inference Provider for Hugging Face, supporting serverless AI model services.