Vercel offers NVIDIA's Nemotron 3.5 Lightning for free
Vercel is opening its AI Gateway to a new open-weight model built for long-running agents, and its memory span sounds almost unreasonable. Here's how it works, and what you might be trading away.
Vercel offers NVIDIA's Nemotron 3.5 Lightning for free
Vercel's AI Gateway now lists NVIDIA's Nemotron 3.5 Lightning as a free model under the name nvidia/nemotron-3.5-lightning-free. NVIDIA serves the model, and it is aimed at long-running agents, sub-agent work, chatbots, and RAG systems.
Nemotron 3.5 Lightning is an open-weight reasoning and chat model. It has 30B parameters, with 3B active per token. Its hybrid Mixture-of-Experts design interleaves Mamba-2 and MoE layers with a few attention layers. The model has a 1M token context window and supports tool calling and structured outputs. The provider may retain prompts for model improvement.
Attribute
Value
Model name
nvidia/nemotron-3.5-lightning-free
Price
Free
Type
Open-weight reasoning and chat model
Parameters
30B total, 3B active per token
Architecture
Hybrid Mixture-of-Experts interleaving Mamba-2 and MoE layers with a few attention layers
Context window
1M tokens
Features
Tool calling and structured outputs
You can call the model through the AI SDK, the Chat Completions, Responses, and Messages APIs, or through a coding agent connected to AI Gateway. With the AI SDK, you pass the model name directly:
import { streamText } from 'ai'; const result = streamText({ model: 'nvidia/nemotron-3.5-lightning-free', prompt: 'Write a TypeScript function that retries a failed fetch with exponential backoff.',});
To use the model in Claude Code, Codex, Cursor, and other agents, install the latest Vercel CLI and run npx vercel ai-gateway setup. Then select nvidia/nemotron-3.5-lightning-free in the agent. Vercel's coding agents guide covers the details.
AI Gateway provides a unified API for calling models and tracking usage and cost. It also handles retries, failover, and performance optimizations aimed at higher-than-provider uptime. Built-in features include custom reporting, budgets for API keys, and routing rules.
Key Features at a Glance
Parameters: 30B total, 3B active per token – a small active footprint for each request, given the model's total size.
Architecture: Hybrid Mixture-of-Experts with Mamba-2 and MoE layers plus a few attention layers.
Context window: 1M tokens – room for long documents and extended agent sessions.
Tool calling and structured outputs: supported – suited to agent workflows that need function calls and predictable output formats.
Multiple APIs: AI SDK, Chat Completions, Responses, and Messages – works with existing client code.
Coding agent support: Claude Code, Codex, Cursor, and more, via npx vercel ai-gateway setup.
Failover and retries: AI Gateway configures these to aim for higher-than-provider uptime.
Availability
The model is free on AI Gateway under the name nvidia/nemotron-3.5-lightning-free. AI Gateway reflects provider pricing with no markup and charges no platform fee on inference, including Bring Your Own Key (BYOK) requests. The article does not give a launch date. Vercel's full list of text models is available on its AI Gateway models page.
