Open models, datasets, Spaces and tools from the Hugging Face Hub and its community, each story linked to its source.
Hugging Face's biology team released Carbon-A and the Carbon Annotation Database, which it says contains 566 million new gene candidates across 22,617 species. Some candidates were validated in the lab with partners.
LightOn published LightOnOCR-3 in 0.8B, 1B and 4B sizes. Besides transcribing text, the models output labeled bounding boxes for document regions, describe images and pull numbers from charts.
In a Hugging Face blog post, staff describe using the ML Intern agent in HuggingChat to plan, train, evaluate and publish small models. One 0.8B model cost about $16 in compute.
Abu Dhabi's Technology Innovation Institute released Falcon-ASR, a 1.6B-parameter speech recognition model. It reports a 20.92% average word error rate across six Arabic test sets.
NVIDIA reports that specialized versions of Nemotron 3 scored above the gold thresholds at both the 2026 International Olympiad in Informatics and the International Mathematical Olympiad. The runs were unofficial.
The llama.cpp server now supports decision models through a new endpoint. These models read an input once and return a probability for each option you give them.
The Allen Institute for AI released AstaBrief 8B, which turns a research question and retrieved literature excerpts into a cited report. It powers the Fast mode of report generation in Ai2's Asta platform.
Hugging Face and collaborators launched the Open TTS Leaderboard to evaluate open-source text-to-speech and voice-cloning models across languages. The authors say open models are underrepresented on arena-style rankings.
Paris-based H Company introduced Holo4, agentic models in 27B dense and 35B-A3B mixture-of-experts sizes, plus an updated Holotron4 Nano. The models work through GUIs, code, MCP and APIs.
Hugging Face added an RL Environments filter to the Hub so agent training and evaluation environments can be shared as dataset repos. It works with frameworks including Harbor, Verifiers, OpenEnv and NVIDIA NeMo Gym.
The ESPnet team published YODAS v3, an open dataset of about 1.1 million hours of real-world speech in more than 100 languages, with 48kHz audio and timestamped transcripts.