Latest stories
MPIE-Bench: New Benchmark Exposes Anatomical Flaws in Multi-Person Image Editing
Hugging Face researchers introduce MPIE-Bench, a 2,500-sample benchmark that reveals persistent anatomical and geometric errors in...
Flux-OPD: New On-Policy Distillation Method Uses Evolving Contexts to Improve Open-Ended LLM Training
Researchers propose Flux-OPD, an on-policy distillation paradigm that leverages evolving contexts as in-training supervision to ca...
Memory Decoder at Scale: 6.9B Parametric Memory Boosts Small Models Past 12B Baselines
Researchers scale parametric long-term memory to 6.9B parameters and 300B tokens, showing that pairing a small backbone with a lar...
VideoCoCo: Using Blender Code as Chain-of-Thought for Physically Consistent Video Generation
Hugging Face researchers introduce VideoCoCo, an agentic dual-engine framework that uses executable Blender code as a process-leve...
Hugging Face Unveils Qwen-UI-Agent: A Real-World-Centric Foundation GUI Agent
Qwen-UI-Agent, a new foundation GUI agent from Hugging Face, unifies mobile, computer, browser, and DeepSearch environments, setti...
BM25 Beats Dense and Agentic RAG at Scale, New Study Finds
A controlled scaling study of retrieval-augmented generation (RAG) reveals a scale-dependent crossover: while agentic search leads...
Beacon: A New AI Model That Knows When to Use Tools for Visual Reasoning
Researchers propose Beacon, a novel agentic visual reasoning model that improves multimodal LLM performance by adaptively invoking...
Hugging Face Unveils PhiZero: A World Model That Reasons in Physical Language
PhiZero, a new physical world model from Hugging Face, uses a compact discrete representation called 'physical language' to reason...
AskChem: A Claim-Centered Search Engine for Chemistry Literature
Hugging Face researchers introduce AskChem, a claim-centered infrastructure that converts chemistry papers into atomic, provenance...
Frontis-MA1: Open-Source AI4AI Model Pushes Recursive Self-Improvement in ML Engineering
Hugging Face researchers introduce Frontis-MA1, a 35B-parameter model trained on the OpenMLE stack to improve machine learning eng...
Hugging Face Introduces Metis, the First Memory Foundation Model with Native Memory Capabilities
Researchers at Hugging Face propose memory foundation models, embedding persistent, evolving memory directly into the model backbo...
Hugging Face Unveils GPT-Red: Self-Play Red Teaming to Fortify GPT-5.6
Hugging Face introduces GPT-Red, an automated red-teaming agent trained via self-play to discover novel prompt injection attacks,...