NVIDIA AI

Meet Nemotron 3 Nano Omni 👋 Our latest addition to the Nemotron family is the highest efficiency, open multimodal model with leading accuracy. 30B parameters. 256K context length. 🧵👇

@adcock_brett Amazing - congrats to the team 👏

This is crazy - 2 hours away from 24 hours of continuous humanoid work! The robots have sorted over 28,000 packages so far Bob, Frank, and Gary are all healthy

@HermesAgentTips 💯

@NVIDIAAI They are all unique in their own way!

💚 seeing these builds

@0xSero Let's gooo!!

My next big thing.

💡 Why did @togethercompute choose NVIDIA Blackwell to serve DeepSeek-V4? Because NVIDIA Blackwell is built for the bottlenecks that matter most in long-context inference: → KV-cache pressure during decode → MoE weight bandwidth during prefill A single NVIDIA HGX B200 system can keep DeepSeek-V4’s compressed CSA/HCA/SWA cache layouts resident across many concurrent long-context requests, while native MXFP4 support enables efficient end-to-end quantized inference for V4’s MoE weights. The result? Higher throughput, lower overhead, and optimized serving efficiency at scale.

Thanks for building with us @CrusoeAI 💚 As context windows explode, tokenization is becoming a major hidden bottleneck in inference pipelines. fastokens is open source, already integrated with Dynamo & @lmsysorg, and designed for the next generation of 100K-token agent systems.

The tokenizer inside @NVIDIA Dynamo? That's fastokens — built by Crusoe. On NVIDIA GB200 NVL72 it delivers 9.1x average speedup over Hugging Face and up to 40% faster TTFT. We don't just run open-source. We build it. Read the full breakdown: crusoe.ai/resources/blog…

@xqybyt 🙌

@NVIDIAAI nemotron 3 super. it's insanely good!

Curious what people are running locally these days 👀

@PrimeIntellect @nvidia @vllm_project @sgl_project Congrats! 👏

We use renderers across Lab, verifiers, and prime-rl. We are collaborating with leading open-source partners, including @NVIDIA @vllm_project @sgl_project, to ensure it can become a useful standard across models, inference engines, and RL infra stacks throughout the ecosystem. primeintellect.ai/blog/renderers

Introducing Renderers RL trainers work in tokens. Environments work in messages. Going back and forth corrupts sampled tokens, wasting compute on every agentic turn. With Renderers, we fix this mismatch. This unlocks >3x throughput on popular open models.

@Recursive_SI Congrats!

@LangChain @nvidia we see you!

Special guests at the Ask me Anything booth. @NVIDIA! Stop by during lunch!

@mouser58907 🙌

Since I got it stable I've burned 1B tokens on Qwen3.6 that would have cost over $500! This thing will definitely pay for itself in under 1 year. No regrets so far. Running Total (22 days): 1,046,312,172 tokens | 15,630 messages | $525.62 @NVIDIAAI

@0xSero Congrats!

I just received a 100,000$ grant from the Human Rights Foundation. In total I received: - 100K USD through HRF - 25.8K USD through donations site - 25K Brev credits through Nvidia - 4x B200s for a month - 5K from lambda - 4x RTX PRO 6000 private donor Open source must win

Read more 👇 nvda.ws/4eKCZhz

Delivering agentic inference at scale requires balancing efficiency across: 1) Models and algorithms 2) Software 3) Compute Our full-stack platform continuously optimizes for these inputs using extreme co-design across compute, networking, storage, and memory. Plus, software with broad ecosystem support across millions of developers. The result: lower cost per token, higher throughput, and more scalable AI systems.

OpenShell v0.0.40 🔀 local-domain service routing in the gateway ☸️ k8s node scheduling + tolerations 🔒 CLI TLS now uses the OS trust store 🛡️ SecretResolver debug no longer leaks secrets Two security fixes ship alongside new routing and K8s scheduling control. github.com/NVIDIA/OpenShe…

@JoeGuglielmucci 🙌

@NVIDIAAI Nemoclaw obviously. It's insane to even be testing anything else.

@sambommakanti awesome!

@NVIDIAAI I just built my home robot to use VSLAM and navigate the environment autonomously. Based on NVIDIA Jetson Orin Nano super dev kit and RGB-D camera. Next to pure voice interaction capability through LLM and NemoClaw to get it to do things and interact with users 💪🏽

Build Video Analytics AI Agents with Skills x.com/i/broadcasts/1…

@sudoingX great to see you using it 💚

nobody is talking about how good nemotron 3 nano omni 30b-a3b actually is on local. very underrated. multimodal, reasoning, video understanding, image vision, all shipped in one open source release by nvidia. moe architecture 30b total params, 3b active per token, q8 is near lossless and fits comfortably on a single dgx spark with room to breathe. i have been running it for weeks now and the gap between what this model can do and what the conversation says is wide. nvidia is pushing hard on the open-source front. most builders haven't noticed yet because the discourse is locked on closed-source frontier benchmarks and the next viral chart. meanwhile this thing handles agentic loops, processes video inputs, reasons across image context, and stays responsive on consumer tier unified memory hardware. on dgx spark it flies. more content coming, showing all the modalities in action. if you have used it, what is your experience. drop your stack and your findings, curious what other builders are seeing across hardware tiers.

@poolsideai @nvidia @PrimeIntellect @huggingface This will be great! Can't wait to see what people build

Poolside is hosting a 2-day model research hackathon in London. Join us to push an open-weight agent model as far as you can. RL and fine-tune Laguna XS.2, our latest-generation model, on Prime Intellect Lab. Dates: May 29–30 Partners: @nvidia + @PrimeIntellect + @huggingface Prize: NVIDIA DGX Spark Agents need better models. Better models need cracked researchers. Link below.