Research
Holo3.1 ships quantized agents that run locally on your laptop
Hugging Face Blog · Jun 2, 2026 · 2 min read
Hcompany just dropped Holo3.1, and it's not a minor point release. This is the first time the company has shipped quant...
AI Research
Cutting-edge AI research papers, breakthroughs, and academic developments.
587 articles
Research
JetBrains drops Mellum2, a 12B open model that runs 2x faster by only waking a fifth of its brain
Hugging Face Blog · Jun 1, 2026 · 2 min read
JetBrains just open-sourced Mellum2, a 12-billion-parameter Mixture-of-Experts model built for the unglamorous, high-fr...
Research
IBM's agents use 30× fewer tokens than pure LLMs by giving AI a map, not just a brain
Hugging Face Blog · Jun 1, 2026 · 2 min read
The narrative around enterprise AI has become a broken record: companies launch pilots, then watch them crash into the...
Research
PyTorch profiling isn't just for experts: Here's how to read your first trace
Hugging Face Blog · May 29, 2026 · 2 min read
Most PyTorch users know they should profile their models. Few actually do it. The barrier isn't a lack of tools — it's...
Research
Async RL's dirty secret: 98% of weights don't change between steps, and TRL now exploits that
Hugging Face Blog · May 27, 2026 · 2 min read
Every async RL library has a dirty secret: every single step, the trainer ships the entire model to the inference engin...
Research
Reachy Mini cuts the cloud entirely with a local AI voice pipeline
Hugging Face Blog · May 27, 2026 · 2 min read
A new tutorial from Hugging Face shows how to run Reachy Mini, the open-source robot, completely offline. No cloud. No...
Research
AI's Tower of Babel: Why 'Harness' and 'Scaffold' Mean Everything and Nothing
Hugging Face Blog · May 25, 2026 · 2 min read
If you've ever nodded along while secretly wondering what separates an LLM 'harness' from its 'scaffold,' you're in cro...
Research
OlmoEarth v1.1 slashes satellite AI compute by 3x by killing redundant tokens
Hugging Face Blog · May 19, 2026 · 2 min read
The team behind OlmoEarth just made a move that's more about smart engineering than raw power, and it could quietly res...
Research
Six new open-source AI rerankers just dropped, hitting state-of-the-art accuracy at every size
Hugging Face Blog · May 19, 2026 · 2 min read
A new family of six open-source cross-encoder rerankers, all built on the ModernBERT architecture, is now available and...
Research
PaddleOCR 3.5 now runs on a Transformers backend, cutting a major friction point for RAG devs
Hugging Face Blog · May 18, 2026 · 2 min read
The team behind PaddleOCR just shipped version 3.5, and the real news isn't a new model—it's a new inference engine opt...
Research
IBM's 97M-param Granite model crushes sub-100M multilingual retrieval at 60.3
Hugging Face Blog · May 14, 2026 · 3 min read
IBM just flipped the script on multilingual embeddings with Granite Multilingual R2. For years, the trade-off was bruta...
Research
Synchronous batching wastes 24% of GPU time: Here's how async cuts it to zero
Hugging Face Blog · May 14, 2026 · 2 min read
If you're running inference on an H200 at $5 an hour, every second the GPU sits idle is money you're lighting on fire....
Research
GPU Math: Why AWS's Latest Instances Won't Fix Your Memory Bottleneck
Hugging Face Blog · May 11, 2026 · 2 min read
If you're training or running inference on large models, you already know the pain. The step time that should be domina...
Research
Four V1 bugs nearly sank our RL training — here’s what vLLM 0.18.1 actually fixed
Hugging Face Blog · May 6, 2026 · 2 min read
Nobody should have to debug a reinforcement learning pipeline by staring at a clip-rate chart that looks like a heart a...
Research
Hugging Face locks down new speech datasets to stop AI benchmark gaming
Hugging Face Blog · May 6, 2026 · 2 min read
The Open ASR Leaderboard just got its first taste of private test data — and it's a direct shot at the subtle art of 'b...
Research
IBM’s Granite 4.1 packs 30B dense model punch using just 15T tokens and a 5-stage data diet
Hugging Face Blog · Apr 29, 2026 · 2 min read
IBM just dropped Granite 4.1, and the playbook they're sharing is a direct challenge to the “scale is all you need” ort...
Research
DeepInfra joins Hugging Face Hub, slashing serverless AI costs for devs
Hugging Face Blog · Apr 29, 2026 · 2 min read
Hugging Face just got a serious cost-performance boost. DeepInfra, the serverless inference platform known for aggressi...
Research
NVIDIA's New 30B Model Is 9x Faster at Parsing Your 100-Page PDFs and Video Archives
Hugging Face Blog · Apr 28, 2026 · 2 min read
NVIDIA is taking a swing at the omni-modal crown with the new Nemotron 3 Nano Omni, and the spec sheet suggests it isn'...
Research
OpenAI's privacy filter runs locally with just 50M active parameters, enabling 128k-token document scans
Hugging Face Blog · Apr 27, 2026 · 3 min read
The Privacy Filter model isn't another massive, resource-hungry beast. OpenAI shipped it as a 1.5B-parameter Apache 2.0...
Research
DeepSeek's V4 slashes KV cache to 2% of normal size, making million-token agents practical
Hugging Face Blog · Apr 24, 2026 · 2 min read
The dirty secret of AI agents is that they break. Not dramatically, but in predictable, boring ways that make long-runn...
Research
Running Local AI in a Chrome Extension: The MV3 Architecture That Actually Works
Hugging Face Blog · Apr 23, 2026 · 2 min read
Building a Chrome extension that runs local AI models isn't just about picking the right library—it's an architectural...
Research
Up to 40% of Arabic AI Benchmarks Are Broken, QIMMA Audit Finds
Hugging Face Blog · Apr 21, 2026 · 2 min read
If you've been tracking Arabic LLM evaluation, you've probably noticed a growing tension: the number of benchmarks and...
Research
Mythos cracked vulns fast — but its real lesson is the system, not the model
Hugging Face Blog · Apr 21, 2026 · 3 min read
The AI model getting all the attention is called Mythos, a frontier LLM that’s been finding and fixing software vulnera...
Research
AI code agents flood open source with 10x more pull requests no one can review
Hugging Face Blog · Apr 16, 2026 · 3 min read
Code agents are working. Jensen Huang says we've gone from 30 million coders to a billion overnight. Anyone with an age...