NVIDIA AI releases & news
Every NVIDIA AI release AI/TLDR has verified from a primary source: 40 so far, mostly 14 models, 8 ecosystem and 6 tools. Latest: Nemotron 3.5 Transcribe — NVIDIA's open speech model you steer with prompts (2026-10-28).
40 releases tracked · www.nvidia.com
- Nemotron 3.5 Transcribe — NVIDIA's open speech model you steer with prompts
One open checkpoint for transcripts, translation, speaker labels and audio Q&A, controlled by plain-language prompts.
- Nemotron 3.5 Super VL — NVIDIA's open 120B vision model for agents
NVIDIA's new open model reads text, documents, charts and short videos in one checkpoint, with only 12B parameters active per token.
- Long-WAM — NVIDIA's robot world-action model remembers 19 seconds of video
A robot policy that watches the last 19 seconds, imagines what comes next, then acts, fast enough for real robots.
- NVIDIA Open Agent Safety Platform — a hardware watchdog for AI agents
Agent safety rules move off the machine the agent runs on and into a separate chip it cannot reach.
- SoL-Pi — NVIDIA's harness extension cuts coding-agent tokens by about half
Four efficiency tricks, picked by an automated research loop out of 152 candidates, packaged as a drop-in Pi extension.
- NVIDIA to acquire Hugging Face — $12.93B, and the hub stays multi-vendor
NVIDIA is buying Hugging Face for $12.93 billion and says the hub will stay open to every model, cloud and chip vendor.
- NVIDIA Groq 3 LPX — the agent inference chip enters full production
NVIDIA's Groq 3 LPX accelerator is in full production, built to decode tokens fast enough to keep AI agents responsive.
- NVIDIA AVO — an agent system scores 100% on ARC-AGI-3's public set
NVIDIA's AVO agent cleared every public ARC-AGI-3 environment, a benchmark where the best single model sat near 30%.
- Nemotron 3.5 Lightning — NVIDIA's 30B open MoE for always-on agents
NVIDIA's new 30B open MoE keeps only 3B parameters active per token, aimed at agents that fire thousands of small steps.
- NeMo Switchyard — NVIDIA's open router picks a model per agent step
An Apache-2.0 Rust proxy from NVIDIA that reshuffles which model handles each step of an agent run.
- NVIDIA Alpamayo 2 Super — 34B open VLA model for robotaxis and self-driving
NVIDIA's largest open vision-language-action model yet, aimed at production robotaxis.
- NVIDIA Agent Toolkit — PhysicsNeMo and CUDA-X plug into agents for chip and physics work
NVIDIA's Agent Toolkit gains PhysicsNeMo and CUDA-X libraries as agent-ready skills, aimed at automating chip design and physics workflows.
- Open Secure AI Alliance — 50-company coalition open-sources AI security tools
A cross-industry alliance to give cyber defenders open AI models and agent frameworks they can inspect, run, and modify.
- NVIDIA ModelExpress — Rust sidecar makes vLLM cold-loads 48× faster
NVIDIA's open-source Rust sidecar loads model weights into GPUs up to 48× faster than a cold Hugging Face pull.
- NVIDIA Cosmos 3 Edge — 4B world model that runs physical AI on Jetson
NVIDIA's Cosmos family gets a small, on-device sibling built for real-time robotics.
- Nemotron 3 Embed — NVIDIA's open 8B embedder takes #1 on RTEB
NVIDIA ships an open 8B embedder that takes #1 on RTEB, plus two efficient 1B variants for production RAG.
- NVIDIA Nemotron 3.5 ASR Streaming 0.6B — Open-Weights Multilingual Speech Model Sustains ~17× More Concurrent Streams on H100, Covers 40 Locales From One Checkpoint
A 600M-param open ASR model that streams 40 locales from one checkpoint and squeezes ~17× more concurrent voice sessions onto an H100.
- NVIDIA Acquires Kumo AI for Over $400M — Predictive Foundation-Model Startup With Reddit, DoorDash, and Walmart on the Customer List Joins NVIDIA, Cofounders Vanja Josifovski, Hema Raghavan, and Stanford's Jure Leskovec Already Inside
NVIDIA buys predictive-foundation-model maker Kumo AI for >$400M to bolt structured-data inference onto its enterprise stack.
- NVIDIA Vera Rubin Hits Full Production — Next-Gen Rack-Scale AI Supercomputer Unifies 72 Rubin GPUs, 36 Vera CPUs, NVLink 6, BlueField-4, and Spectrum-X for 3.6 Exaflops NVFP4 Inference, 20.7 TB HBM4, and 10× Throughput Per Megawatt Over Grace Blackwell
NVIDIA's Vera Rubin AI factory platform is now in volume production, with NVL72 racks shipping fall 2026.
- NVIDIA Nemotron 3 Ultra — 550B Mamba-Transformer Mixture-of-Experts With 55B Active Parameters Tops US Open-Weights Leaderboard at 48 on Artificial Analysis Intelligence Index, Ships June 4 on Hugging Face, OpenRouter, ModelScope, and build.nvidia.com
NVIDIA's biggest open-weights model: a 550B Mamba-Transformer MoE that beats every other US open model on intelligence and inference speed.
- NVIDIA RTX Spark Unveiled at Computex 2026 — Arm-Based Superchip Pairs a 20-Core Grace CPU With a 6,144-Core Blackwell GPU, Up to 128 GB Unified Memory, and 1 Petaflop of FP4 Compute for Slim Windows Laptops
NVIDIA finally builds its own PC chip — a Grace Arm CPU bonded to a Blackwell RTX GPU with 128 GB of shared memory and a petaflop of FP4 throughput.
- NVIDIA Cosmos 3 — Open Mixture-of-Transformers Omni-Model for Physical AI Lands at Computex With Nano (16B) and Super (65B) Variants, OpenMDW 1.1 Weights, and a New Coalition With Runway, Black Forest Labs, and Skild AI
An open frontier omni-model that can reason about, simulate, and act in the physical world.
- NVIDIA LocateAnything-3B — Parallel Box Decoding Vision-Language Grounder Hits 12.7 Boxes/Sec on H100, ~10× Faster Than Qwen3-VL, Trained on 138M Queries Across 785M Boxes
LocateAnything decodes full bounding boxes in one shot, getting NVIDIA a vision-language grounder that is roughly 10× faster than Qwen3-VL.
- NVIDIA's Gamma-World — Generative Multi-Agent World Model Scales Beyond Two Players With Simplex Rotary Agent Encoding and Sparse Hub Attention, Streams Shared Worlds at 24 FPS
NVIDIA's Gamma-World scales generative world models past the two-player ceiling with linear-cost agent attention and 24 FPS real-time rollouts.
- NVIDIA Reports Record $81.6B Q1 FY27 — Data Center Hits $75.2B, Q2 Guide Lifted to $91B, $80B Buyback Authorization, and CEO Jensen Huang Says Vera CPU Opens a 'Brand New $200B' Agentic-AI Market
NVIDIA's blow-out Q1 print pairs $81.6B revenue with a $200B Vera CPU TAM call from Jensen Huang.
- LongLive-2.0 — NVIDIA's NVFP4 Parallel Infrastructure Generates Minute-Long Interactive Video at 45.7 FPS
A 4-bit training and inference stack that makes minute-long interactive video generation fast enough for real time.
- SANA-WM — NVIDIA's 2.6B Open-Source World Model Generates 720p One-Minute Video With 6-DoF Camera Control
An efficient open-source world model that generates minute-long, camera-controllable 720p video.
- NVIDIA AnyFlow — Any-Step Video Diffusion Distillation With Flow-Map Transition Learning Beats Consistency Methods at 1.3B and 14B Scale
Train one video diffusion student that runs well at 1, 4, or 32 steps — no separate model per step budget.
- NVIDIA's 2026 AI Equity Bets Top $40B — $30B in OpenAI, $3.2B Corning, $2.1B IREN, ~24 Private Rounds
TC and CNBC tally NVIDIA's 2026 AI equity bets — $30B in OpenAI, $3.2B in Corning, $2.1B in IREN, ~24 private rounds.
- NVIDIA-IREN $2.1B Partnership Targets 5 GW of AI Infrastructure, Plus $3.4B Cloud Services Deal
NVIDIA stakes IREN with up to $2.1B and signs a $3.4B GPU-cloud contract to build 5 GW of AI factories, flagshipped in Sweetwater, Texas.
- Nemotron-Personas-Korea — 7M Synthetic Personas from Official Korean Demographics
NVIDIA's Korean-language persona dataset: 7 million synthetic personas grounded in official government statistics, CC BY 4.0.
- NVIDIA Ising — Open AI Models for Quantum Processor Calibration and Error Correction
NVIDIA's Ising family applies AI to two core quantum computing bottlenecks: calibrating noisy qubits and decoding quantum errors in real time.
- RoboLab — NVIDIA's 120-Task Sim Benchmark Shows SOTA Robot Policies Top Out at 25.8%
NVIDIA's new 120-task simulation benchmark reveals that the best open robot policies succeed less than 26% of the time on manipulation tasks.
- NVIDIA Nemotron OCR v2 — Unified Multilingual OCR, 28x Faster Than PaddleOCR
A single unified OCR model from NVIDIA that reads six languages at 34.7 pages per second — no per-language models needed.
- Lyra 2.0 — NVIDIA's Explorable Generative 3D World Framework
NVIDIA's open-source framework for generating interactive 3D worlds — walk through, explore, export to Isaac Sim.
- NVIDIA Agent Toolkit — Enterprise AI Agent Platform
NVIDIA's open platform for building enterprise AI agents, with 17 major software companies already on board.
- TriAttention — trigonometric KV cache compression
NVIDIA's new attention technique compresses the KV cache by exploiting a geometric property of Q/K vectors before RoPE is applied.
- PersonaPlex — NVIDIA's real-time speech-to-speech model with persona control
A real-time speech-in/speech-out model from NVIDIA you can prompt into a specific persona and voice, with CPU offload for low-VRAM machines.
- Garak — LLM Vulnerability Scanner by NVIDIA
Automated red-teaming for LLMs — finds vulnerabilities before attackers do.
- NeMo Guardrails — Programmable LLM Safety Rails
Define what your LLM can and can't do using a simple dialogue DSL.