DeepInfra hosts NVIDIA Nemotron-3 Ultra

// 1h agoMODEL RELEASE

DeepInfra hosts NVIDIA Nemotron-3 Ultra

DeepInfra has announced serverless hosting support for NVIDIA's 550-billion-parameter Nemotron-3 Ultra Mixture-of-Experts model. The integration delivers inference speeds exceeding 300 tokens per second for complex reasoning and enterprise-grade agentic workflows.

// ANALYSIS

DeepInfra's rapid adoption of Nemotron-3 Ultra underscores the intense competition among AI inference hosts to serve frontier-level open-weights models at scale. By hosting this 550B parameter giant, DeepInfra democratizes access to enterprise-grade reasoning that would otherwise be cost-prohibitive and complex for individual developers to self-host.

* High-speed, high-density MoE inference showcases DeepInfra's engineering prowess, achieving over 300 tokens per second.

* The model's focus on long-running, autonomous agents signals a major industry shift from simple chat interfaces to complex, multi-step problem-solving systems.

* Hybrid architectures like Mamba-Transformer are proving to be highly competitive and resource-efficient against pure Transformer implementations at this scale.

// TAGS

deepinfranvidianemotron-3-ultramodel-releasellmmoeinferencecloudinfrastructureagentopen-weights

DISCOVERED

1h ago

2026-06-01

PUBLISHED

1h ago

2026-06-01

RELEVANCE

8/ 10

AUTHOR

DeepInfra

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

OPEN SOURCE54m ago

parakeet.cpp delivers 2x faster local ASR

Developed by the LocalAI team, parakeet.cpp is a dependency-free C++17 inference engine for NVIDIA's NeMo Parakeet ASR models that runs up to 2x faster than standard baselines. By leveraging the ggml library to eliminate Python runtime dependencies, it enables highly portable offline speech recognition across CPUs and multiple GPU backends.

MODEL1h ago

NVIDIA has launched Nemotron 3 Ultra, a massive 550-billion parameter hybrid Mamba-Transformer model that is five times faster and 30% cheaper than comparable open models.

NVIDIA has unveiled Nemotron 3 Ultra, the flagship model in its new Nemotron 3 family of foundation models built specifically for complex reasoning, planning, and multi-step agentic workflows. Featuring 550 billion total parameters, the model is powered by a hybrid Mamba-Transformer Mixture-of-Experts (MoE) architecture that utilizes LatentMoE token compression and NVFP4 training to activate only 55 billion parameters per token. This efficient design enables throughputs exceeding 300 tokens per second, a 1-million token context window, and up to 5x faster inference at 30% lower cost compared to existing open-weights models in its class.

VIDEO1h ago

Socket Firewall blocks dependency malware in real time by intercepting package manager network requests at install time.

Socket Firewall is a free command-line interface (CLI) tool designed to proactively protect developer environments from malicious software supply chain attacks. By acting as a local network proxy or command prefix, the tool intercepts package manager network requests (supporting npm, pip, and cargo) during installation, evaluating dependencies in real-time against Socket's security API. This allows it to automatically block known malware, flag suspicious packages with risky capabilities (like unexpected telemetry, network, or filesystem access), and enforce safety policies without disrupting developer workflows.