#NVIDIA

Discover NVIDIA's Orchestrator-8B, enhancing tool selection using reinforcement learning.

NVIDIA's TiDAR combines one-step diffusion drafting with autoregressive verification in a single forward pass to exploit free GPU token slots and multiply tokens-per-forward by up to about 6x while preserving benchmark quality.

'NVIDIA's Reinforcement Learning Pretraining (RLP) injects dense, position-wise reinforcement into pretraining by rewarding chains-of-thought for information gain, improving reasoning and data efficiency across architectures.'

'NVIDIA released ViPE, an open-source Video Pose Engine that converts unconstrained video into camera intrinsics, precise poses, and metric depth maps at scale, backed by massive annotated datasets.'

'UDR by NVIDIA separates strategy from model, compiling user-defined research workflows into sandboxed, auditable execution with LLMs used only for localized reasoning.'

'NVIDIA introduced Jetson Thor, a high‑performance module and developer kit that brings server‑grade multimodal inference and generative reasoning to robotics at the edge. The platform pairs a Blackwell GPU with a focused software stack to enable real‑time perception, planning and action.'

'NVIDIA's Streaming Sortformer enables millisecond, GPU-accelerated speaker diarization for up to four concurrent speakers, producing frame-level labels and timestamps for live transcripts and voice applications.'

'NVIDIA's Nemotron Nano 2 delivers hybrid Mamba-Transformer LLMs that run up to 6× faster and support 128K-token context on a single A10G GPU, with most training data and recipes open-sourced.'

ProRLv2 scales RL training to 3,000 steps and combines regularization and exploration techniques to expand reasoning capabilities in compact LLMs, showing strong benchmark gains across math, coding, logic and STEM tasks.

'NVIDIA announced Cosmos, a full-stack physical AI platform featuring reasoning models, synthetic data tools and Omniverse simulation upgrades to speed robotics development.'

NVIDIA’s XGBoost 3.0 now supports training GBDT models on terabyte-scale datasets using a single Grace Hopper Superchip, delivering massive speed and cost advantages for enterprises.

NVIDIA's ThinkAct framework introduces a dual-system approach combining vision-language reasoning with reinforced visual latent planning, significantly improving robot manipulation and planning in complex environments.

NVIDIA introduces GraspGen, a diffusion-based framework that significantly improves 6-DOF robotic grasping using large-scale synthetic data and innovative training, achieving superior performance in simulation and real-world tests.

NVIDIA's Canary-Qwen-2.5B model sets a new benchmark in speech recognition with a record low Word Error Rate and fast processing speed. This open-source, commercially licensed hybrid ASR-LLM model enables advanced audio transcription and language understanding.

NVIDIA has released Audio Flamingo 3, an open-source model that advances how AI understands and reasons about sound across speech, ambient noise, and music for extended audio durations.

NVIDIA's DiffusionRenderer introduces a groundbreaking AI framework for editable, photorealistic 3D scene generation and manipulation from a single video, unlocking new possibilities in video editing and relighting.

NVIDIA researchers developed Dynamic Memory Sparsification (DMS), a novel method that compresses KV caches by 8× in Transformer-based LLMs, improving inference efficiency while maintaining accuracy.

NVIDIA introduces ProRL, a novel reinforcement learning method that extends training duration to unlock new reasoning capabilities in AI models, achieving superior performance across multiple reasoning benchmarks.

NVIDIA releases Llama Nemotron Nano VL, a compact vision-language model that excels in complex document understanding with efficient multimodal processing and state-of-the-art accuracy.

NVIDIA introduces Llama Nemotron Nano 4B, a compact open-source AI model optimized for edge deployment that outperforms larger models in scientific reasoning and programming tasks.

NVIDIA introduces Cosmos-Reason1, a new suite of AI models designed to enhance physical common sense and embodied reasoning using multimodal learning and innovative ontologies, improving AI interaction in real-world environments.

NVIDIA's Joey Conway discusses groundbreaking open-source AI models Llama Nemotron Ultra and Parakeet, highlighting innovations in reasoning control, data curation, and rapid speech recognition.

NVIDIA has released its Open Code Reasoning models (32B, 14B, 7B) as open-source under Apache 2.0, delivering top-tier performance in code reasoning tasks and broad compatibility with popular AI frameworks.

NVIDIA has released Parakeet TDT 0.6B, an open-source ASR model that transcribes an hour of audio in just one second while achieving top accuracy benchmarks, setting a new industry standard.

NVIDIA introduces Describe Anything 3B, a multimodal large language model that excels in detailed, region-specific captioning for images and videos, outperforming existing models on multiple benchmarks.

NVIDIA has issued an urgent hotfix to address overheating and temperature monitoring problems caused by their recent GPU driver update 576.02, which affected AI and gaming users worldwide.

President Trump's new tariffs have triggered a massive market downturn, heavily affecting AI-related tech stocks and global supply chains. The AI industry's borderless nature faces unprecedented challenges amid escalating trade tensions.