The keynote is over. Let’s talk about what it means for developers. Join NVIDIA experts and the community for an after-hours conversation from GTC Berlin, featuring first reactions, a look at the day’s announcements, and perspectives on open models and agent safety. Hear what’s generating excitement across the event, test your knowledge in an interactive quiz, and enjoy a few community moments along the way. Bring your questions and help shape the discussion about what comes next.
About us
Explore the latest breakthroughs made possible with AI. From deep learning model training and large-scale inference to enhancing operational efficiencies and customer experience, discover how AI is driving innovation and redefining the way organizations operate across industries.
- Website
-
https://developer.nvidia.com/blog/
External link for NVIDIA AI
- Industry
- Computer Hardware Manufacturing
- Company size
- 10,001+ employees
- Headquarters
- Santa Clara, CA
Updates
-
Nemotron 3.5 Super VL, a multimodal open model built to process and reason across text and images, purpose-built for autonomous agents running high-volume, specialized tasks, is available now. In this Ask the Experts session, you’ll get direct access to the NVIDIA AI researchers who built it: the architecture choices, training techniques, and design principles behind a model optimized to run across DGX Spark to the data center, with open weights and recipes you can customize for your domain. What you’ll learn: - How NVIDIA AI researchers trained Nemotron 3.5 Super VL for agentic and vision-language workloads - What the open weights, training recipes, and datasets make possible for domain- and task-specific customization - How Nemotron 3.5 Super VL is optimized to run locally on hardware like DGX Spark without sacrificing accuracy or speed - How to integrate Nemotron 3.5 Super VL into multi-model agentic pipelines Have questions about training, architecture, or running Nemotron 3.5 Super VL in your stack? Ask them live, and our researchers will answer them directly.
Ask the Experts: Nemotron 3.5 Super VL | Nemotron Labs
www.linkedin.com
-
Meet the developers shaping Europe’s AI future. Join us for highlights from developer events across Barcelona, Paris, London, Stockholm, and Berlin - from Build-a-Claw demos and community gatherings to conversations with NVIDIA experts and local builders. Explore how developers are working with AI agents, open models, and physical AI, hear their questions and experiences, and get a preview of the themes heading into GTC Berlin. Bring your questions and join the conversation live.
Road to GTC Berlin: Europe’s AI Builders in the Spotlight
www.linkedin.com
-
Welcome Beam! Congrats to the Reflection team on launching their first open-weight model. The 501B MoE model was trained on NVIDIA GB300 NVL72 and delivers up to 90K tokens/sec/GPU on Blackwell Ultra out of the box on day 0. Beam 1.0 NVFP4 checkpoint on Hugging Face: https://lnkd.in/gBxspJ8Z
Beam is now available. Open weights are now on Hugging Face, under the Apache 2.0 license. With this release, we’re launching what we’re calling an open operating system for agents - the model is the kernel, and everything above the infrastructure layer is open: reference agentic integrations, deployment containers, inference integrations, and customization pipelines. 40+ partners are contributing to our open ecosystem from day one, so every builder has everything they need to deploy, customize, and run Beam in their own environment from the start. Learn more here: https://lnkd.in/gMNnQgxw
-
-
Can’t stop, won’t stop. The open models keep dropping FLUX 3 Image Klein 5.6B was trained on NVIDIA GB200 NVL72 and can run on an RTX GPU. Big 🙌 to the Black Forest Labs team. There's also an NVFP4 variant: https://lnkd.in/gnaRqb49
FLUX 3 Image Klein is here. Open Weights with the quality of a large frontier model. Customizable and fast. 5.6B parameters and runs on a single consumer GPU. Native 4K, bounding boxes, and pixel-perfect editing. In our internal evals, FLUX 3 Image Klein scores higher than the current leading open weights model in text-to-image and multi-reference generation, while costing less. What we’re launching: • Base and distilled open weights for fine-tuning, research and self-hosting. • FP8, NVFP4 and INT8 versions, small enough to run on GPUs and laptops with 8 GB of VRAM • BFL API with native 4K, bounding boxes, and pixel-perfect editing • Launch support for Ostris’s AI Toolkit, ComfyUI, Diffusers, SGLang Diffusion and vLLM-Omni, plus NVIDIA fine-tuning recipes and deployment optimizations Trained on NVIDIA GB200 NVL72: 0.55s per image on NVIDIA GB300, ~4s on an NVIDIA GeForce RTX 5090. For enterprise licensing, custom fine-tuning and deployment support, get in touch. Links in the comments.
-
We have another one today! NVIDIA Nemotron 3.5 Bolt 24GB puts powerful local agents within reach ⚡ 74B-A9B parameters packed into 24GB of memory, optimizing intelligence and speed per GB of memory while running alongside your apps. This hybrid Mamba-Transformer model was compressed from Nemotron 3.5 Super VL by using structured compression, quantized to 2-bit with quantization-aware distillation, and released with an efficient LiLiCorr speculative drafter. As the #1 small model on PinchBench with 84 TPS on RTX Spark, it’s ready for your next local project. Model: https://lnkd.in/gShCqDCw Tech blog: https://lnkd.in/g8EMhtzB Tech report: https://lnkd.in/gq_DM2Q8
-
-
And AI at Meta went live with Muse Spark Open! The 816B model was trained on NVIDIA GB300 NVL72. Its 1M-token context window gives coding agents room to work across large codebases. Meta also produced NVFP4 checkpoints optimized for Blackwell Ultra, delivering over 50K tokens/sec/GPU to improve throughput and lower cost per token. Congrats to the Muse team. More in our blog: https://nvda.ws/4hUJw9h NVFP4: https://lnkd.in/gm-rtewY
Today we’re releasing Muse Spark Open, the most capable US open-weight model to date. Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning. Key Model Specs: → 816B total / 42B active parameters → 1M token context window → Natively multimodal across text, image, video, audio To help make advanced models more accessible, Muse Spark Open is optimized to run on two Apple M5 Ultra Mac Studios with the latest MLX framework. It also runs on leading inference engines like SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, llama.cpp, and Unsloth with optimizations for AMD, Apple, Intel, and NVIDIA silicon. Developers can further customize the model with PyTorch, including TorchTitan for fine-tuning. We’re releasing the weights under the Muse Model License, with limited commercial terms that let many developers use it for free. Learn more on our research blog: https://lnkd.in/eUiykJ6Q Download the weights: https://lnkd.in/ef-SzyDs
-
-
Meet NVIDIA Nemotron 3.5 Super VL 👀 Our new open model combines reasoning and tool use with text, image, document, and video understanding for long-running AI agents. The model has 120B total parameters, with 12B active per token. An agent can read a report, interpret its charts, and search the web for more information using the same model. That reduces the need to pass context between separate text and vision models. Teams are already post-training Nemotron 3.5 Super VL for tasks in document intelligence, cybersecurity, finance, scientific experimentation, and travel. Read how we built it: https://nvda.ws/4jUHF71 Download the model on Hugging Face: https://lnkd.in/e5Q3aS9i
-
Congrats to StepFun on Step 5 Preview 🙌 Their largest open weights release. The 600B MoE model is built to handle agent tasks lasting up to 24 hours. The model runs at nearly 8K tokens/sec/GPU on NVIDIA GB300 using NVFP4. NVP4 checkpoint on Hugging Face: https://lnkd.in/gTBdUvfy Deploy with SGLang - https://lnkd.in/gCafnwS9 or vLLM - https://lnkd.in/gxkubgsF Post-train with NeMo Automodel: https://lnkd.in/g6E5X77p Run it with NVIDIA NIM using a single command: https://lnkd.in/gMgybNJ8
Today we’re releasing Step 5 Preview as open weights. An API lets developers call a model. Open weights let them bring it into their own stack. By opening the weights, improving our developer tools and building with the community, we want more developers to build their own agents, applications and workflows on Step 5 Preview. What’s included: • BF16 and FP8 model weights, on Hugging Face and ModelScope • Technical blog: “Inside Step 5 Preview: Architecture and Training for Agentic Work” • Step Code v0.1.4, our open-source agent harness • Deployment guide, with support for vLLM, SGLang and Transformers across multiple chip platforms About the model: Step 5 Preview is designed for long-horizon agentic tasks. It uses a sparse MoE architecture with 600B total parameters and 27B active per token, supports a 1M-token context window, and natively takes text and vision input. It performs especially well in AI coding, software engineering, professional knowledge work and finance. Since launching on OpenRouter, it has ranked #2 globally on both the daily and weekly usage leaderboards. Inside the technical blog: • A 92-layer narrow-and-deep architecture for implicit multi-hop reasoning • Sparse GQA, which reduces indexing and top-k overhead to 1/8 for million-token contexts • More stable training with FP32 on key states and fine-grained gradient clipping • Million-scale verifiable environments across software engineering, scientific reasoning, ML R&D and professional knowledge work • RL for long-horizon tasks, with context compression built into end-to-end policy training Day-0 ecosystem support: • NVIDIA: try it on build.nvidia.com, fine-tune with NeMo AutoModel, deploy with model-free NIM, and run on Blackwell • Inference platforms: available on Fireworks AI, Together AI, Modal, DeepInfra and Featherless AI Thank you to all our partners for making it available to developers from day one. We’d love to see what you build, and to hear your feedback. Frontier intelligence, open weights. Build what’s next. 🤗 Weights: https://lnkd.in/ggY__w8B / ModelScope: https://lnkd.in/eTG388K3 📄 Technical blog: https://lnkd.in/g_hs3dA3 📘 Deployment guide: https://lnkd.in/gEUe8kWm 💻 Step Code: https://lnkd.in/gj_aV7p9
-