AI Briefing#huggingface
The latest AI and developer news about #huggingface, with the original source and a short summary.
The latest AI and developer news about #huggingface, with the original source and a short summary.
Hugging Face has released ml-intern, a tool for agents that controls ML libraries using local models.
Resemble AI has released DramaBox, a voice synthesis model based on LTX 2.3.
SenseNova has released U1 A3B, a multimodal model based on NEO-unify.
Ovis2.6-80B-A3B has been released as an MoE-based multimodal model.
Hugging Face has released 'physics-intern,' a multi-agent framework for theoretical physics research.
The number of GGUF-format model uploads on Hugging Face has roughly doubled over the past two months.
A fake OpenAI Privacy Filter on Hugging Face reportedly distributed a Rust infostealer.
Hundreds of malicious models and skills have been identified in the Hugging Face and ClawHub repositories.
A Hugging Face model impersonating a legitimate one has reportedly been found downloading malicious scripts.
Hugging Face's `Open-OSS/privacy-filter` has been reported as malware targeting Windows.
Zyphra has released ZAYA1-8B, an 8B MoE model trained on AMD Instinct MI300.
An analysis of Hugging Face user data found that AI builders tend to prioritize VRAM capacity over compute performance.
Hugging Face has released an interactive guide comparing the performance and scalability of various RL frameworks.
z-lab has released the DFlash version of Gemma 4 31B-it on Hugging Face.
DharmaOCR released specialized 3B and 7B SLMs along with cost and performance benchmarks.
IBM has unveiled the Granite 4.1 model series, optimized for multilingual generation, coding, RAG, and AI assistant workflows.
Mistral has released a unified multimodal model with 128B parameters and a 256k context.
A bug in Gemma 4's chat template that failed to handle complex JSON Schema has been fixed.
DeepInfra has been added to Hugging Face's Inference Providers.
Xiaomi MiMo has open-sourced MiMo V2.5-Pro, a 1.02T MoE model.
A 36B MoE model built with Darwin V7 achieved 88.4% on GPQA Diamond.
DeepSeek's latest models, the V4 Flash and Non-Flash versions, have been released via Hugging Face.
vLLM Recipes has been revamped to support hardware-specific serving configs and command generation.
An uncensored Aggressive version of Qwen3.6-27B and K_P quantizations have been released.