Nemotron 3.5 Lightning Is Now Free on AIHubMix

#ai #llm #agents #api

NVIDIA Nemotron 3.5 Lightning is now available to use for free on AIHubMix.

Built for the execution layer of AI agents

Long-running agents spend much of their time on frequent operational work: calling tools, validating outputs, writing and reviewing code, formatting results, and delegating tasks to sub-agents. Using a large frontier model for every one of those steps can add unnecessary latency and cost.

Nemotron 3.5 Lightning is designed for this execution layer. It is an open 30B Mixture-of-Experts model with only 3B active parameters, combining the capacity of a larger model with a much smaller active compute footprint.

Key capabilities include:

30B total parameters with 3B active parameters
Context windows of up to 1 million tokens
Tool calling and tool-output validation
Coding and specialized task execution
Long-running autonomous-agent workflows
Fast, high-volume inference
Local deployment on supported NVIDIA hardware

Lightning for execution, larger models for planning

Nemotron 3.5 Lightning is the smallest member of the Nemotron 3 family. While larger models such as Nemotron 3 Super and Ultra are suited to advanced reasoning, orchestration, and complex planning, Lightning focuses on the high-frequency execution steps that dominate an agent's workload.

That makes it a practical workhorse for sub-agents, coding assistants, tool-driven workflows, and always-on agent systems.

Start using it for free

The free route is currently available on AIHubMix with zero input and output token pricing:

Model ID: nemotron-3.5-lightning-free
Input: $0 / 1M tokens
Output: $0 / 1M tokens
Context: up to 1M tokens

It works through AIHubMix's OpenAI-compatible API:

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["AIHUBMIX_API_KEY"],
    base_url="https://aihubmix.com/v1",
)

response = client.chat.completions.create(
    model="nemotron-3.5-lightning-free",
    messages=[
        {"role": "user", "content": "Help me validate this tool output."}
    ],
)

print(response.choices[0].message.content)