NVIDIA puts a 30B reasoning model on AI Gateway for free: 3B active per token, 1M context, built for long-running agents

NVIDIA has placed Nemotron 3.5 Lightning on AI Gateway, offered free under the model name nvidia/nemotron-3.5-lightning-free and served by NVIDIA. The model was not previously among the language models available through AI Gateway, so callers had to find another entry point. AI Gateway provides a unified API for calling models, tracking usage and cost, and configuring retries and failover; the model now appears in its list of available language models.

It is an open-weight reasoning and chat model intended for long-running agents, sub-agent work, chatbots, and RAG systems. It uses a hybrid Mixture-of-Experts design that interleaves Mamba-2 and MoE layers with a few attention layers, rather than relying on attention layers throughout. The model has 30B parameters with 3B active per token, a 1M token context window, and support for tool calling and structured outputs.

Calls can go through the AI SDK, the Chat Completions, Responses, and Messages APIs, or a coding agent connected to AI Gateway. Inside coding agents such as Claude Code, Codex, and Cursor, use requires installing the latest Vercel CLI, running npx vercel ai-gateway setup, and then selecting the model name in the agent. AI Gateway reflects provider pricing with no markup and charges no platform fee on inference, including on Bring Your Own Key requests.

The report covers the model's free availability on AI Gateway, its specifications, and how to call it, for developers building long-context agents and RAG systems. The source states that the provider may retain prompts for model improvement, which is relevant when deciding what data to submit.

The source notes that the provider may retain prompts for model improvement, a condition to keep in mind when using the free endpoint; beyond that, the source gives no further account of the model's capability limits.