NVIDIA Nemotron 3.5 Lightning is an open mixture-of-experts model from NVIDIA, with 3B active parameters out of 30B total. It is suited for high-throughput agentic workloads and specialized tasks that benefit from domain-specific customization. A free-tier endpoint is also available on OpenRouter.

NVIDIA's open reasoning and orchestration model, 55B active parameters out of 550B on a hybrid Transformer-Mamba mixture of experts. Text in, text out, with a context window up to 1M tokens. Built for long-running agent work: orchestration, coding agents, deep research and multi-step reasoning.

A 4B multimodal guardrail model from NVIDIA, fine-tuned from Gemma-3-4B. It moderates both the input to an LLM and the response, taking text and images and returning a safe or unsafe classification for each, category labels, and an optional reasoning trace. 12 languages, 128K context.

NVIDIA's Nemotron 3 Super, a 120B open hybrid mixture-of-experts model activating 12B parameters per token. The Mamba-Transformer backbone with multi-token prediction generates over 50% more tokens than comparable open models, and Latent MoE calls 4 experts for the inference cost of one. 1M token context. Weights, datasets and recipes are published under the NVIDIA Open License.

NVIDIA-Nemotron-Nano-9B-v2 is a large language model (LLM) trained from scratch by NVIDIA, designed as a unified model for reasoning and non-reasoning tasks. It can expose an internal reasoning trace and then produce a final answer, or be configured via system prompt to only provide final answers without intermediate traces.

NVIDIA Nemotron 3.5 Lightning is an open mixture-of-experts model from NVIDIA, with 3B active parameters out of 30B total. It is suited for high-throughput agentic workloads and specialized tasks that benefit from domain-specific customization. A free-tier endpoint is also available on OpenRouter.

NVIDIA's open reasoning and orchestration model, 55B active parameters out of 550B on a hybrid Transformer-Mamba mixture of experts. Text in, text out, with a context window up to 1M tokens. Built for long-running agent work: orchestration, coding agents, deep research and multi-step reasoning.

A 4B multimodal guardrail model from NVIDIA, fine-tuned from Gemma-3-4B. It moderates both the input to an LLM and the response, taking text and images and returning a safe or unsafe classification for each, category labels, and an optional reasoning trace. 12 languages, 128K context.

NVIDIA's Nemotron 3 Super, a 120B open hybrid mixture-of-experts model activating 12B parameters per token. The Mamba-Transformer backbone with multi-token prediction generates over 50% more tokens than comparable open models, and Latent MoE calls 4 experts for the inference cost of one. 1M token context. Weights, datasets and recipes are published under the NVIDIA Open License.

NVIDIA-Nemotron-Nano-9B-v2 is a large language model (LLM) trained from scratch by NVIDIA, designed as a unified model for reasoning and non-reasoning tasks. It can expose an internal reasoning trace and then produce a final answer, or be configured via system prompt to only provide final answers without intermediate traces.