NVIDIA drops Nemotron 3 Ultra

// 2h agoMODEL RELEASE

NVIDIA drops Nemotron 3 Ultra

NVIDIA Nemotron 3 Ultra is a 550-billion parameter mixture-of-experts model optimized for agentic workflows and tool calling. Built on a hybrid Transformer-Mamba architecture, the model supports a 1-million token context window and offers up to 5x faster inference.

// ANALYSIS

NVIDIA is successfully transitioning from a hardware provider to a leading AI model powerhouse by directly solving the efficiency constraints of pure Transformer architectures for agentic systems.

* The hybrid Transformer-Mamba architecture allows the model to process up to 1 million tokens with linear scaling, significantly cutting inference costs.

* Granular reasoning budgets enable dynamic computational scaling, giving developers fine-grained control over execution latency and accuracy.

* The release of a 550B parameter model optimized for NVFP4 quantization lowers the barrier for enterprise self-hosting.

// TAGS

nvidianemotronmambatransformermoereasoningagentic-workflowsmodel-release

DISCOVERED

2h ago

2026-06-04

PUBLISHED

2h ago

2026-06-04

RELEVANCE

9/ 10

AUTHOR

Prompt Engineering

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

NEWS10m ago

Anthropic's Claude models demonstrate rapid acceleration in recursive self-improvement, with the Mythos Preview model achieving a 52x speedup on optimization tasks.

Anthropic's latest models have shown dramatic progress in recursive self-improvement (RSI) capabilities. According to internal reports, Anthropic tasks newly released models with optimizing the training code for smaller AI models. While Claude Opus 4 averaged a 3x speedup in May 2024, the newly developed Mythos Preview model achieved a 52x speedup in April 2026, demonstrating that AI-driven self-optimization is accelerating at an exponential rate.

UPDATE26m ago

ChatGPT rolls out background memory system

OpenAI is rolling out a new background memory system for ChatGPT Plus and Pro users in the US that doubles capacity and automatically curates memories using broader chat history via a process called "dreaming." Users retain full control with the ability to manage saved memories through a new dashboard or revert to the legacy memory experience in settings.

LAUNCH41m ago

ElevenLabs has introduced Flows Agent in ElevenCreative, a conversational assistant that automatically builds and iterates node-based, multi-modal creative workflows.

ElevenLabs has introduced the Flows Agent within its ElevenCreative platform, a tool that allows creators to build and modify complete creative workflows using natural language. The agent handles tasks such as selecting models, creating nodes, wiring connections, and running generations across over 50 image, video, voice, music, and sound effects models. With an active assist mode, users maintain cost control by approving expensive operations, while the system supports background processing so workflows can complete even after closing the tab. Users can iterate on their pipelines dynamically through conversation—such as swapping voices, backgrounds, or languages—without rebuilding the entire flow from scratch.