NV-Embed-v2 is NVIDIA's 2024 embedding model for retrieval with a context of around 32768 tokens, kept as a catalog snapshot.

Llama 3.1 Nemotron 70B Instruct is NVIDIA's RLHF-tuned Llama 3.1 70B, reporting Arena Hard 85.0 and MT-Bench 8.98, with 128K input context.

NVIDIA's efficient open-weight 30B MoE hybrid model with 3B active parameters, 1M-token context, and single-GPU deployment for local reasoning and coding.

Mistral Nemo 12B (open-mistral-nemo-2407) is an Apache 2.0 model with 128k context; Mistral deprecated it May 22, 2026 and retired it July 31, 2026.

NVIDIA nv-rerankqa-mistral-4b-v3 is a 2024 question-answering reranker for NeMo Retriever NIM, with a 512-token limit and a CIKM 2024 workshop paper.

NVIDIA's open-source agent framework that turns an agent into one Python class: methods are actions, docstrings are prompts, at 82.2% SWE-bench Verified.

NVIDIA OpenShell runs AI agents in policy-controlled sandboxes with filesystem restrictions, network rules, and protected provider credentials.

SoL-Pi is NVIDIA's MIT Pi extension that cuts agent tokens with four opt-in mechanisms and no patch to Pi itself.

NVIDIA's open-source LLM vulnerability scanner with dozens of plugins and thousands of prompts for probing jailbreaks, prompt injection, and data leakage.

GMI Cloud is an AI-native GPU cloud for training, fine-tuning, and inference, offering on-demand NVIDIA H100, H200, and Blackwell GPUs with hourly billing.

NVIDIA NeMo Guardrails is an open-source toolkit for adding programmable guardrails to LLM chat apps via Colang input, dialog, retrieval, and output rails.

NVIDIA CUDA MCP server and Nsight Copilot for coding agents. Hosted docs MCP plus an Apache 2.0 self-hosted blueprint.

Orukeet is a 25-language ASR model that freezes fitted Gabor kernels on NVIDIA Parakeet TDT 0.6B v3.