NVIDIA AI

Teaching your AI new tricks.

[NVIDIA AI @NVIDIAAI]

Today we're shipping Nemotron 3 Ultra.

A 550B MoE frontier-intelligence open model built for long-running agents.

It delivers 5x faster inference and lowers the cost of complex agentic tasks by up to 30% versus other open frontier models.

[NVIDIA AI @NVIDIAAI]

@QuantCapitalX @MiniMax_AI https://t.co/o0McAL0tnM

[NVIDIA AI @NVIDIAAI]

Congrats to the @MiniMax_AI team on the release of MiniMax M3, a long-context multimodal model for text, image, and video reasoning. 🙌

Try it today with our free GPU-accelerated endpoint on https://t.co/es07MrU5I0.

Details: https://t.co/89qlcTP3OW

[MiniMax (official) @MiniMax_AI]

MiniMax M3, Open-Weight, Now On Hugging Face , with only ~428B parameters and ~23B activated parameters

Weights:
https://t.co/g4Ybfa2kWH
MiniMax Sparse Attention:
https://t.co/HcTlWRotG3

[NVIDIA AI @NVIDIAAI]

@refikanadol Always love seeing your installations! Huge 🙌 to your team.

[Chris 🇨🇦 @llm_wizard]

So you wann run Nemotron 3 Ultra, and you have 4 DGX Sparks?

Well, have I got a resource for you!

Huge thanks to @eugr for helping put this guide together so people can play with Ultra with semi-reasonable tok/s!

[NVIDIA AI @NVIDIAAI]

@GergelyOrosz Great having you at HQ!

[NVIDIA AI @NVIDIAAI]

@ivanfioravanti 🙌

[NVIDIA AI @NVIDIAAI]

We're proud to join @linuxfoundation and industry peers in launching Akrites.

"Transparency and open collaboration are how the cybersecurity community has kept infrastructure safe for decades. In the age of AI, these open source foundations have never been more critical." — David Reber, Chief Security Officer, NVIDIA

[The Linux Foundation @linuxfoundation]

The age of AI requires a new solution to cybersecurity challenges. Today, the Linux Foundation and industry leaders launched Akrites, a coordinated effort to defend critical open source software against AI-enabled cyber threats.

Frontier AI can find software vulnerabilities in minutes. Akrites introduces a shared SIRT and a standardized, confidential disclosure process to remediate vulnerabilities upstream before they are exploited.

Read the open letter and learn why top organizations are joining together in this critical effort
https://t.co/sgyZ7Gajec

[rishit dagli @rishit_dagli]

What if we can take a few photos and turn it into an interactive physically realistic virtual world🌎?

📢Introducing AdaVoMP (ICML 26): generates volumetric physics fields at high spatial resolution making objects interactive and deformable🦾

🦣16^3x higher res (1024^3)
⚡️more accurate

[NVIDIA AI @NVIDIAAI]

@SlimTradeyBaby @UnslothAI 😂

[NVIDIA AI @NVIDIAAI]

@togethercompute @ArtificialAnlys @FeelTheBeurn 🔥

[NVIDIA AI @NVIDIAAI]

@SlimTradeyBaby @FrostForger Love to hear this!

[NVIDIA AI @NVIDIAAI]

@TheAhmadOsman 🦾

[NVIDIA AI @NVIDIAAI]

@rheapatel_ @nvidia @github @Microsoft 🙌

[NVIDIA AI @NVIDIAAI]

@kylekuzma @Meta And let us know when you're in town next - we'd love to have you come over to our HQ.

[NVIDIA AI @NVIDIAAI]

@arsh_goyal Yup, not efficient.

[NVIDIA AI @NVIDIAAI]

@gmi_cloud @VoiceCursor @snaplii 🙌

[NVIDIA AI @NVIDIAAI]

@alexmashrabov Sweet - what NVIDIA tech are you using?

[NVIDIA AI @NVIDIAAI]

@ivanfioravanti Nice! They're a little piece of art.

[NVIDIA AI @NVIDIAAI]

@VenelinVidenov @SlimTradeyBaby Nice!! What are you building?

[NVIDIA Robotics @NVIDIARobotics]

Cosmos 3 wins another leaderboard. The power of an open frontier model. 🎉

Ranked #1 on @allen_ai's MolmoSpaces robot policies leaderboard, a policy model post-trained on Cosmos 3 Nano shows the strongest performance in environments that require models to interpret, reason, and act in real time.

Not just understanding the world, but interacting with it.

📊 https://t.co/FPWArD2PmE

Drop what you're building in the comments 👇

[NVIDIA AI @NVIDIAAI]

@SlimTradeyBaby @huggingface 💚

[NVIDIA AI @NVIDIAAI]

The rise of MoE models introduced new challenges in training, and @huggingface's Transformers v5 brought first-class support for solving them.

Now, NeMo AutoModel builds on top of v5. Part of the NeMo framework for building models at scale, NeMo AutoModel brings optimizations to a broad set of model families through support for Expert Parallelism, DeepEP, and TransformerEngine kernels with a few lines of code.

We found NeMo AutoModel brings a 3.4 to 3.7x higher training throughput for popular MoE models. You can read more here: https://t.co/TNlBsKWwrJ