Nemotron 3.5 Lightning
Neural Network
Nemotron 3.5 Lightning is an open NVIDIA model with a mixture of experts architecture for high-performance agentic tasks.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 1 000 000 tokens
- Max answer length
- 65 536 tokens
- Reasoning effort
- low, medium, high
- Capabilities
- reasoning, function calling
How it works Nemotron 3.5 Lightning?
Nemotron 3.5 Lightning is an open NVIDIA model built on a mixture of experts architecture: out of 30 billion parameters, about 3 billion active ones are used at each step. This allows the model to maintain high throughput, making it suitable for agentic workloads with high call volumes and specialized tasks. In BotHub, it is accessible without a VPN or foreign card: pay with a local card in rubles only for actual usage, and unused tokens do not expire. Over 250 models are available in one window, allowing you to switch and compare responses without rewriting integrations: the API is OpenAI-compatible, and chats are protected by AES-GCM encryption. Companies can work via contract with invoicing and EDI, using an admin panel and team limits. It is useful if you are building agentic pipelines where request stream processing speed is critical, performing mass classification and data labeling, extracting structured fields from documents, or using a fast layer in a multi-model system while reserving heavy models for complex steps.Frequently asked questions about 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 million-token context: it analyzes large documents, helps with code—refactoring, debugging, log parsing—and solves tasks requiring reasoning and calculations. It accepts images and files as input and supports function calling, making it suitable for agentic scenarios.
The single response limit is 65,536 tokens. This is enough for a long article, a large code module, or a detailed analysis of an entire document without splitting it into parts. When working via API, you set your own output limit and use the remaining context for input materials.
Yes, reasoning before answering is supported: the model first outlines the solution path and then provides the result. This is noticeable in math, logic problems, code debugging, and analyzing contradictory documents. For short questions, this mode is usually not needed—the answer comes immediately.
Yes, the model supports function calling and structured JSON output. You describe the tools, it selects the right one and returns arguments in a parsable format—convenient for agents, document parsing, and integration with your backend. In BotHub, this works via a unified OpenAI-compatible API.
Images and documents—yes: you can send a screenshot, diagram, chart, or file and ask to analyze the content. Input support for audio and video is not indicated in our data, so for voiceovers or video analysis, it is better to use a specialized model—in BotHub, it opens in the same window.
The quality of Russian is not specified in our data, and we cannot guarantee it. Test it with your own task: formulate a typical query and compare the result with another model—in BotHub, switching takes a couple of seconds, and you only pay for the tokens actually consumed.