NVIDIA Nemotron 3.5 Lightning
Neural Network
Nemotron 3.5 Lightning is an open NVIDIA MoE model with 3B active parameters: Nemotron API for agentic tasks and high loads
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 262 144 tokens
- Max answer length
- 235 929 tokens
- Reasoning effort
- low, medium, high
- Capabilities
- reasoning, function calling
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works NVIDIA Nemotron 3.5 Lightning?
Nemotron 3.5 Lightning is an open NVIDIA model based on the mixture-of-experts architecture: out of 30 billion parameters, about 3 billion are activated per request, allowing the model to maintain high throughput for agentic workloads and specialized tasks. Via BotHub, it is accessible without a VPN or foreign card: pay with a Russian card in rubles for tokens used, no subscription required, and unused tokens do not expire. Access over 250 models in one window, and a unified OpenAI-compatible API allows switching between them without rewriting integrations; traffic is encrypted via AES-GCM, data is not stored, and teams have access to contracts, invoices, EDI, an admin panel, and limits. Use Lightning for multi-step agents and call chains where response speed is critical, for streaming processing of repetitive requests like classification, labeling, and data extraction, for internal assistants and service prototypes, and for load testing your prompts before deployment.Frequently asked questions about NVIDIA Nemotron 3.5 Lightning
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
This is an NVIDIA text model with a 262,144 token context. It is suitable for analyzing long documents, writing and refactoring code, debugging, reasoning, and STEM tasks, as well as reading images and building agentic scenarios with tool calling.
Up to 131,072 tokens per response. This is enough for voluminous technical documentation, large code modules, detailed research analysis, or an extensive report that doesn't need to be manually assembled from multiple separate requests.
Yes, reasoning mode is confirmed. The model can build intermediate steps before the final answer, which helps in mathematics, logic, data analysis, and parsing complex code where the sequence of calculations is important, rather than a quick guess.
Yes, both function calling and structured JSON output are supported. The model can be connected to external services and databases, and the predictable response format simplifies data parsing on your application side without additional parsers.
Images and documents are accepted as input: you can ask it to analyze a diagram, interface screenshot, chart, or multi-page file. There is no mention of audio or video in our data. The model returns the result as text.
The model is multilingual, and you can work with Russian-language queries in BotHub just like any others. It is easy to check the quality on your task: open a chat, send a typical prompt, and if necessary, compare the result with another model in the same window.