---
title: "Best Use of Nemotron – Junction 2026"
description: "Build a project powered by NVIDIA Nemotron that solves a real problem through reasoning, tool use, or coordinated workflows. Applies to: Any track."
url: https://junction2026.com/challenges/nvidia-best-use-of-nemotron
---

> For the index of all pages on this site, see [llms.txt](https://junction2026.com/llms.txt).

# Best Use of Nemotron

Junction 2026 · Tech challenges

by NVIDIA

Build a project powered by NVIDIA Nemotron that solves a real problem through reasoning, tool use, or coordinated workflows. Applies to: Any track.

#### INTRODUCTION

What can you build when AI moves beyond answering questions and starts getting things done? NVIDIA invites you to use Nemotron to build a project that tackles a real problem for a clear audience. Connect it to tools, ground it in useful data, or combine it with other models to turn an idea into a working solution. Show us what Nemotron enables and why it matters to the people using your project.

#### THE TECHNOLOGY

Meet Nemotron 3.5 Super VL, NVIDIA's newest open model, and it's yours to build with this weekend. It has 120B parameters but only wakes up 12B per token, so it runs fast for its size, and it can read text, images, documents and even video in a single workflow. That means your agent can look at a screenshot, read a PDF, watch a clip, call a tool and write the code to act on it, all with one model.

#### THE CHALLENGE

Build a project where Nemotron plays a meaningful role in doing real work. Use it on its own or alongside other models, tools, and data sources to solve a problem for a clear audience. Nemotron must contribute directly to the project's core functionality, and your demo should make that contribution visible. The solution is open-ended: choose the problem, design the workflow, and show what your approach enables.

**Minimum Requirements:**

- Use NVIDIA Nemotron 3.5 Super VL model in the working solution.
- Demonstrate its contribution to the project's core functionality.
- Submit a short explanation identifying the model used, what it does, why it matters, and how you improved its outputs.
  - This should be done as text input into the submission form.

#### GETTING STARTED

- **NVIDIA Build:** Free hosted APIs to explore, test, and prototype with NVIDIA and community AI models.
  - Note: The API is OpenAI-compatible with base URL <https://integrate.api.nvidia.com/v1> (for example, model `"nvidia/nemotron-3.5-lightning-30b-a3b"`). The free tier is rate-limited (developer forums report about 40 requests per minute per key), so agent loops should use retries with backoff and each teammate should use their own key.
- **OpenRouter API:** Serverless path with a unified API across providers.
  - Note: Free variants (ending in `:free`) are capped at 20 requests per minute and 50 requests per day, or 1,000 per day once the account has bought at least 10 credits.
- **Hugging Face:** Open weights for local runs, fine-tuning or deployment on Brev.

#### RESOURCES AND SUPPORT

- **Support Channel:** Nvidia Discord channel and at the NVIDIA table

#### PRIZES AND JUDGING

#### PRIZE

#### 2 winners of 1x RTX 5080: JUDGING CRITERIA

- **Meaningful Nemotron Usage:** Nemotron is central to the project's value. The team clearly explains what it does and why it matters.
- **Technical Execution:** The demo works reliably and shows thoughtful implementation choices around architecture, data flow, tool use, latency, or error handling.
- **Quality of AI Output:** Outputs are useful, relevant, and grounded where appropriate. The team demonstrates efforts to improve quality through prompting, evaluation, or feedback loops.
- **Creativity and Differentiation:** The team uses Nemotron thoughtfully or inventively, with a distinct approach to the problem.

#### COMPANY AND CONTACTS

##### About NVIDIA

NVIDIA builds accelerated computing technologies and AI software that help developers turn ideas into applications. At Junction, we're inviting teams to explore NVIDIA Nemotron open models and build useful agents around real-world needs. Start with a problem you care about and show what you can do over the weekend.

##### Contacts

- **Main Contact:** Elliot Winch, Community Program Manager, EMEA, <ewinch@nvidia.com>
- **Technical Contact:**
- **Mentors Attending:** Maxime Pariente and Ryan Reese
- **Judges:** NVIDIA Team

## Challenge partner

NVIDIA

Tech challenges

## Company Info: NVIDIA

Website: <https://www.nvidia.com/en-us/>

## Judging Criteria

- Meaningful Nemotron Usage — (25%)

  Nemotron is central to the project's value. The team clearly explains what it does and why it matters.

- Technical Execution — (25%)

  The demo works reliably and shows thoughtful implementation choices around architecture, data flow, tool use, latency, or error handling.

- Quality of AI Output — (25%)

  Outputs are useful, relevant, and grounded where appropriate. The team demonstrates efforts to improve quality through prompting, evaluation, or feedback loops.

- Creativity and Differentiation — (25%)

  The team uses Nemotron thoughtfully or inventively, with a distinct approach to the problem.

## Prizes

- Two best teams

  1x NVIDIA RTX 5080