How much does it cost to customize Nemotron 3.5 Lightning?

According to Vice President of Generative AI Kari Briski, one partner (CodeRabbit Inc.) trained a router agent using Nvidia's standard auto model recipe for $85 in around two hours. Another partner set up a training job on a single H100 card overnight with minimal cost.

What is NeMo Switchyard and how does it work?

NeMo Switchyard is an open-source model routing library that takes into account available AI models and their capabilities, then routes prompts to the most capable and efficient model for each step of an agent workflow. Developers can customize the router with their own strategy based on priorities like quality, latency, or cost.

Can Nemotron 3.5 Lightning be customized with a company's own data?

Yes. Lightning is open and customizable, meaning it can be readily post-trained with Nvidia NeMo on an enterprise's own hardware using proprietary domain data, tools and workflows to improve accuracy for specialized tasks. Nvidia is also releasing its post-training datasets and training recipes so developers can blend them with their own data.

Large Language ModelsOpen-Source AIAI Business & IndustrySiliconANGLE AIAug 12, 2026Aug 12, 2026, 19:01 JST

Nvidia releases Lightning model and AI router to let enterprises pick the right AI for each task

Send on LINE

3 Key Points

What happened
Nvidia announced Nemotron 3.5 Lightning, a 30 billion-parameter open model designed for high-volume agentic AI tasks, and NeMo Switchyard, an open-source routing library that directs prompts to the most capable model for each step of an agent workflow. Lightning delivers up to four times the output speed and 30% faster agentic task completion compared with other models in its weight class, and can be customized on enterprise hardware with proprietary data.
Why it matters
As enterprises deploy AI agents that must handle tasks of varying complexity—from simple sorting to deep reasoning—no single model fits all purposes. Lightning's low cost and ease of customization (one partner trained a router agent for $85 in two hours) let companies tailor models to their own domain data without expensive infrastructure. Switchyard lets them route each task to the most efficient model available, cutting costs and improving accuracy instead of defaulting to a single expensive model for all work.
What to watch
Nvidia is releasing its post-training datasets and the recipes used to train the models, allowing developers to blend Nvidia's data with their own. The company is collaborating with partners including Boomi LP, Cadence Design Systems Inc., Classmethod Inc., Cognition AI Inc., Kong Inc., Langchain Inc., Nous Research Inc. and Siemens AG on intelligent model routing integration.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

Nvidia's announcement reflects a fundamental shift in enterprise AI deployment: the era of picking a single "best" model for all tasks is ending. As AI agents handle diverse workloads—some requiring speed and efficiency for simple classification, others demanding reasoning depth—enterprises face a cost and accuracy optimization problem that no monolithic model can solve. Lightning addresses the customization bottleneck by making fine-tuning cheap and fast enough to run on modest hardware; Switchyard addresses the routing problem by letting developers orchestrate multiple models based on real-time task requirements.

The strategy leverages Nvidia's position in enterprise infrastructure. By open-sourcing both the model and the routing framework, and by releasing training datasets and recipes, Nvidia increases adoption friction for competitors while building a de facto standard for multi-model orchestration. The early partner examples—CodeRabbit training a router for $85, another partner integrating Lightning with zero pipeline changes—suggest the barrier to entry is genuinely low. This matters because it shifts the competitive advantage from raw model capability to the software stack that lets enterprises extract value from multiple models efficiently.

FAQ

How much does it cost to customize Nemotron 3.5 Lightning?: According to Vice President of Generative AI Kari Briski, one partner (CodeRabbit Inc.) trained a router agent using Nvidia's standard auto model recipe for $85 in around two hours. Another partner set up a training job on a single H100 card overnight with minimal cost.
What is NeMo Switchyard and how does it work?: NeMo Switchyard is an open-source model routing library that takes into account available AI models and their capabilities, then routes prompts to the most capable and efficient model for each step of an agent workflow. Developers can customize the router with their own strategy based on priorities like quality, latency, or cost.
Can Nemotron 3.5 Lightning be customized with a company's own data?: Yes. Lightning is open and customizable, meaning it can be readily post-trained with Nvidia NeMo on an enterprise's own hardware using proprietary domain data, tools and workflows to improve accuracy for specialized tasks. Nvidia is also releasing its post-training datasets and training recipes so developers can blend them with their own data.

SiliconANGLE AIRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.