Sponsored by Byond Boundrys Consulting - Empowering Ideas, Delivering Results
NVIDIA Advanced

NVIDIA Nemotron 3.5 Lightning 30B A3B

NVIDIA Nemotron 3.5 Lightning 30B A3B is an open-weight 30B (3B active) MoE for fast sub-agents and long-running agents, with up to 1M context and a free prototype endpoint.

Reasoning ModelsText Freemium
In plain English

What is this model and why does it matter?

NVIDIA Nemotron 3.5 Lightning 30B A3B is a 30B-total, 3B-active MoE model released in August 2026 for fast long-running autonomous agents and sub-agent workloads with up to 1M context.

SubagentsLong-running agentsCodingRAGCost-efficient private inference
Model overview

NVIDIA Nemotron 3.5 Lightning 30B A3B: features, use cases and important details

NVIDIA Nemotron 3.5 Lightning 30B A3B is a current AI model verified from first-party NVIDIA sources.

NVIDIA Nemotron 3.5 Lightning 30B A3B verified specifications

NVIDIA Nemotron 3.5 Lightning 30B A3B is a 30B-total, 3B-active MoE model released in August 2026 for fast long-running autonomous agents and sub-agent workloads with up to 1M context. Its verified context or usage limit is Up to 1000000 tokens, with Deployment configurable; official hosted examples use 16384 tokens maximum output.

NVIDIA Nemotron 3.5 Lightning 30B A3B pricing and access

NVIDIA offers a free prototype endpoint and downloadable weights for Nemotron 3.5 Lightning. Production cost depends on partner endpoint or self-hosted GPU infrastructure.

NVIDIA Nemotron 3.5 Lightning 30B A3B best uses

Subagents, Long-running agents, Coding, RAG, Cost-efficient private inference.

NVIDIA Nemotron 3.5 Lightning 30B A3B limitations

Open-weight does not mean zero infrastructure cost, Performance depends on serving stack, Use-case validation is required.

Get started

How to use this model

  1. Generate an NVIDIA API key or deploy the NIM locally.
  2. Call nvidia/nemotron-3.5-lightning-30b-a3b.
  3. Use recommended sampling.
  4. Enable reasoning for harder tasks.
  5. Test throughput and agent reliability on your workload.
Copy and try

Example prompts

  • Act as a fast subagent and inspect these files.
  • Handle this long-running coding task with periodic checkpoints.
  • Reason over this knowledge base and call the appropriate tools.
Capabilities

What it can do

  • 30B total / 3B active
  • 1M context
  • MoE
  • Reasoning
  • Tool use
  • Downloadable weights
  • Speculative decoding support
Best for

Practical use cases

  • Coding subagents
  • Agent swarms
  • RAG
  • Private AI
  • High-throughput automation
Pricing

What does it cost?

NVIDIA offers a free prototype endpoint and downloadable weights for Nemotron 3.5 Lightning. Production cost depends on partner endpoint or self-hosted GPU infrastructure.

InputFree prototype; production infrastructure dependent
OutputFree prototype; production infrastructure dependent
Simple summaryPrototype API use is free; production deployment cost depends on provider or GPU infrastructure.

What stands out

  • Very new August 2026 model
  • Small active parameter count
  • 1M context
  • Free prototype endpoint

Things to consider

  • Custom OpenMDW license
  • Text-only
  • Production GPU tuning required
Limitations

Important restrictions and trade-offs

  • Open-weight does not mean zero infrastructure cost
  • Performance depends on serving stack
  • Use-case validation is required
SimplifyAITools verdict

Our editorial take

A timely high-demand NVIDIA model for developers looking for a fast, efficient workhorse inside multi-agent systems.

References

Primary sources

  1. Open source 1 ↗
  2. Open source 2 ↗
Keep exploring

Related AI models