📣 Announcing the release of OpenReasoning-Nemotron: a suite of reasoning-capable LLMs which have been distilled from the DeepSeek R1 0528 671B model. Trained on a massive, high-quality dataset distilled from the new DeepSeek R1 0528, our new 7B, 14B, and 32B models achieve SOTA perf on a wide range of reasoning benchmarks for their respective sizes in the domain of mathematics, science and code. The models are available on @huggingface🤗:

🎶 Meet Audio-Flamingo 3 – a fully open LALM trained on sound, speech, and music datasets. 🎶 Handles 10-min audio, long-form text, and voice conversations. Perfect for audio QA, dialog, and reasoning. On @huggingface ➡️ From #NVIDIAResearch.

🏆 Our NVIDIA NeMo Canary-Qwen-2.5B scores 1st place on the @HuggingFace leaderboard for automatic speech recognition, setting a new record with a 5.63% word error rate. 🤗 Leaderboard ➡️

What if you could ask a chatbot a question the size of an entire encyclopedia—and get an answer in real time? Multi-million token queries with 32x more users are now possible with Helix Parallelism, an innovation by #NVIDIAResearch that drives inference at huge scale. 🔗

Python + OpenUSD = faster workflows, easier automation, more creativity. 💡 Python’s scripting capabilities and ease of use make it the perfect partner for #OpenUSD, helping accelerate development without needing C++ expertise. Learn more 👉

✨ FLUX.1 Kontext brings a fresh experience to image generation through incremental, in-context editing rather than traditional complex prompts. Now optimized for NVIDIA RTX GPUs using TensorRT, we’ve collaborated with @bfl_ml to deliver faster inference with lower VRAM requirements. 🔗 Details:

Deploy your own DeepSeek-R1 model instance with 1-click deployment. The new NVIDIA TensorRT-LLM container on Vertex AI is optimzed for LLMs at scale and significantly improves performance and cost-efficiency. Get started today ➡️

📸 Generate. Refine. Repeat. @bfl_ml’s FLUX.1 Kontext simplifies AI image editing with text + image prompts—now RTX-accelerated with TensorRT. Bonus: Google’s new Gemma 3n multimodal LLM is also accelerated on RTX. 🔗 #RTXAIGarage:

Ready to unlock the power of #multimodal retrieval? Discover how NVIDIA #NeMoRetriever transforms document search—combining vision and language for smarter, faster, more accurate results. 👉

🏆NVIDIA NeMo Retriever achieves 1st place on @HuggingFace across three top visual document retrieval leaderboards -- ViDoRe V1, ViDoRe V2, and MTEB VisualDocumentRetrieval. Setting new standards for enterprise #RAG applications, NeMo Retriever leads the industry. This top-performing model is: 🔎 Fine-tuned for query-document retrieval where queries are text and images. 📊 Particularly useful in multimodal RAG systems that use text, charts, tables or infographics. Potential applications include multimedia search, cross-modal retrieval, and conversational AI with rich input understanding. 🛠️ Developers can achieve up to 15X faster PDF data extraction, 50% better accuracy, and 35X better storage efficiency with NeMo Retriever. For researchers, check it out on #HuggingFace ➡️ For developers needing production-ready, commercial models, visit ➡️