Language generation beyond one token at a time. 💬
NVIDIA Nemotron-Labs-Sigma generates blocks of tokens in parallel, opening the door to faster generation for exploring multiple solutions to math and coding problems.
Read the paper: arxiv.org/abs/2610.02665
Try the open models: huggingface.co/collections/n…
📣 NVIDIA is releasing a collection of open model weights: Nemotron-Labs-Sigma — 14 checkpoints at 3B/8B.
Language models have stayed discrete. We bring continuous diffusion to language as in vision.
Smooth ODE trajectories (16-D) unlock distillation and steering: 8 TPFs for math, 4 TPFs for code, with accuracy and diversity preserved. 🧵 1/
NVIDIA post-trained Nemotron with Factory, running entirely within its own air-gapped network.
Factory was the only commercially supported harness deployed by @nvidia in their environment, giving researchers access to long-running agents and full execution traces.
🧪 Start with a driving clip, a robot task, a store scene, or a circuit board.
NVIDIA Cosmos Data Factory blueprints generate variations, including rare events and defects, check the results, and prepare training data.
One prompt sets up the environment; a second runs your chosen workflow.
Read the technical walkthrough → nvda.ws/4AWkheX
Watch @NVIDIA, @Voxel51 and @Encord build training-ready datasets with Cosmos Data Factory on Nov. 19 at 11 AM. PT. Bring your questions for the live Q&A.
Set a reminder ⬇️
youtube.com/watch?v=c5or9gzM…
2
2,863
Accelerate Visual AI Agents on NVIDIA Jetson with VSS 3.3 and Cosmos www.teapawt.org/i/broadcasts/2mdqbjBOr…
Beyond Transcription: Nemotron 3.5 Transcribe | Nemotron Labs www.teapawt.org/i/broadcasts/2AMLjZMgB…
Introducing NVIDIA Nemotron for Science, just announced at #SC26 by Ian Buck, our VP of Hyperscale and High-Performance Computing.
The collection includes open models, datasets, tools, and post-training recipes to help researchers build more accurate and efficient AI agents for science. Agents can plan tasks, run simulations, check results, and adjust as they go.
New walkthrough shows how to post-train Nemotron 3.5 Super VL for science using NeMo Gym and NeMo RL ⬇️
nvda.ws/493WJIN
From detecting assembly defects to flagging missed steps in real time, @MercedesBenz is using AI agents powered by @NVIDIA Cosmos to improve production quality.
Michael Schiebe, Board Member at Mercedes-Benz, shares how physical AI is helping build self-learning factories where machines adapt and work alongside people.
20 million downloads for NVIDIA Cosmos on @HuggingFace! 🎉
To the developers and researchers building robots, autonomous vehicles and vision AI agents with open world models: thank you for experimenting, sharing feedback and building with us.
Tell us what you're building with Cosmos next! 💚
Check out NVIDIA Cosmos on Hugging Face ⬇️
huggingface.co/collections/n…
Nemotron 3.5 Transcribe takes #1 on @ArtificialAnlys’ new Diarization Index! 💚
Thanks to the team for testing how well models identify who said what in conversations with 2–8 speakers.
Announcing the Artificial Analysis Diarization Index, our new benchmark for who said what in speech to text
The Diarization Index scores APIs on Word Diarization Error Rate (WDER): the share of words given to the wrong speaker or to no speaker, across 97 conversations of two to eight people, with equal weight on each group size.
Nemotron 3.5 Transcribe from @NVIDIAAI leads at 3.3%, ahead of Voxtral Small Transcribe 3 at 7.5%. Thread below 🧵
See a recorded driving scene from a viewpoint the vehicle never captured.
NVIDIA Research’s LidarVerse is the first model to simulate scans for different LiDAR sensors, which use lasers to capture surroundings in 3D. It can recreate scenes from new viewpoints, generate scenes from maps, and predict what a sensor will see next, all in real time.
Read the paper and get the code ➡️ bit.ly/4zIQki0
Learn how AI is helping build smarter cities ⬇️
youtube.com/watch?v=ha-gDrDm…
Congrats @thinkymachines team on Inkling 2 Small!
A 594B model pretrained in just 68 hours with NVIDIA NVFP4 on GB300 NVL72 racks.
Read how the team made low-precision training work at this scale. 👇
Training with fewer bits is faster, but quantization error can interfere with learning.
Today on Connectionism, we detail how we trained Inkling 2 Small with a native 4-bit forward pass from the first step of pretraining through the final RL rollout.
thinkingmachines.ai/blog/end…
Download the Inkling 2 Small NVFP4 check point from @HuggingFace: huggingface.co/thinkingmachi…
Meet Déjà View, a model that turns photos taken from different angles into 3D geometry and camera positions.
By looping a single transformer block, this 117M-parameter model matches or outperforms models up to 10x larger across five benchmarks, using comparable or less compute. Researchers can dial the number of loops to fit their compute budget.
#NeurIPS2026 oral paper + code: research.nvidia.com/labs/dvl…
Try Déjà View with the new interactive demo, now available on @HuggingFace with no coding required: huggingface.co/spaces/nvidia…








