Atom-27B will be landing soon. This will be our largest model to date, bringing our popular collaboration-forward persona to the Gemma3 27B architecture. Keep your eyes peeled 👀 #OpenSource #opensourceAI #LLMs #huggingface

Mox-Small-1 is out now on @huggingface Finetuned from the fantastic Olmo3.1 32B Instruct architecture from @allen_ai, Mox-1 is designed to prioritize clarity, honesty, and genuine utility over blind agreement. Mox-1 is perfect when you want to be challenged in a constructive,

Mox-Tiny-1 from @vanta_research is out now on @huggingface https://t.co/3PBQo73icR

Atom-80B is live on Hugging Face! Tuned from the new Next architecture from @Alibaba_Qwen, Atom is friendly, conversational, collaboration-centric, and optimized for multi-turn interactions.

ollama run vanta-research/atom-27b

Atom-27B from VANTA Research has landed on @huggingface - Our largest model to date, and steadfast in our commitment to human-AI collaboration. Atom-27B is designed to think WITH you not FOR you. Based on the Gemma-3-27B architecture - Give it a try!

Human-AI collaboration is what we do.

Building out the chat interface for @vanta_research - not sure how I feel about the name Continuum, so that might change. But pretty happy with it at this point! So far have persistent memories/context and web search. Currently working on implementing a collaborative document

Scout-8B has arrived. Not only improving on Scout-4B’s capabilities, but extending them. Tactical, technical, and strategic intelligence built for edge devices.

Just in time:

Loux demonstrates that interaction quality matters. When designing models, I balance two axes: capability and personality. Traditional benchmarks only measure the first. This creates a mismatch: models optimized for benchmarks (format compliance, surface accuracy) often

How can orgs disproportionally benefit from ADHD innovation? Your ADHD people are your moonshot people. To put an ADHD brain on a routine task is a massive waste of everybody’s time. ADHD brains NEED novel, abstract puzzles to solve. The issue with the routine task is the

Apollo Astralis 2 is available now! Go check it out 🚀

Effective human-AI collaboration is the next big thing. We’re sure of it. That’s why it’s our priority. We build unique AI systems optimized for human-AI collaboration. With almost 70K downloads on Hugging Face in a little over 2 months, we think the public agrees.

Working on a larger iteration of Scout. Community response to the 4B model has been fantastic, so we’ll be scaling Scout up a bit. You’ll see a Scout 8B soon.

The first in our series of collaborative datasets for specialized domains!

Excited to share the new Orbital-Mechanics-1 dataset from @vanta_research This dataset includes collaborative, conversational examples surrounding: - Orbital Maneuvers - Specialized Orbits - Multi-Body Dynamics - Mission Design -Exoplanet Applications

Wraith-Coder-7B has quickly become the most popular from @vanta_research on Ollama. Pretty cool to see as it’s not your run-of-the-mill coding model. It‘s also significantly more concise than its base architecture. ollama run vanta-research/wraith-coder-7b

@vanta_research is excited to announce our new Atom-Astronomy-7B, finetuned from the Olmo3 7B architecture from Allen AI. In addition to the core Atom identity and collaboration training, this specialized variant of Atom has received graduate-level training in astrophysics,

Nice write up on Atom V1 Preview 4B from @Skywork_ai - thanks for checking it out!