---
title: "Use NVIDIA GPT-OSS 20B"
description: "Connect DrowAI to NVIDIA's hosted GPT-OSS 20B endpoint for free development and prototyping."
section: "How To"
updated: 2026-07-21
tags: ["nvidia","gpt-oss","models","api","setup"]
source: "/user-guide/how-to/nvidia-gpt-oss-integration"
---

# Use NVIDIA GPT-OSS 20B

Connect DrowAI to NVIDIA's hosted GPT-OSS 20B endpoint for free development and prototyping.

NVIDIA provides a hosted endpoint for `openai/gpt-oss-20b` that can be used for
free development and prototyping, subject to NVIDIA's account requirements,
trial terms, and usage limits. This is a convenient way to test DrowAI without
starting with a paid model provider.

## 1. Open the NVIDIA model page

Go to the
[NVIDIA Build page for `openai/gpt-oss-20b`](https://build.nvidia.com/openai/gpt-oss-20b).
The page identifies the model and provides its Build, Playground, Model Card,
and API Reference views.

## 2. Generate an NVIDIA API key

After signing in and accepting the applicable terms, return to the **Build**
view and select **Generate API Key**. Copy the generated key to a secure
password manager or secret store.

The integration uses NVIDIA's OpenAI-compatible API endpoint and the model ID
`openai/gpt-oss-20b`. DrowAI supplies those values automatically; you only need
to provide the NVIDIA API key.

## 3. Add the key in DrowAI

1. Sign in to DrowAI with an administrator account.
2. Open the account menu and select **Settings**.
3. Stay on the **API** tab and scroll to **Open models**.
4. In the **NVIDIA** card, paste the key into **API Key**.
5. Select **Update**.

## 4. Select GPT-OSS 20B in Operations

Return to **Operations** and open the model selector in the Command Post. Choose
**Open models**, then **GPT-OSS 20B**. The selector displays **GPT-OSS 20B /
NVIDIA** when the NVIDIA-hosted model is selected.

Create or select a task and send a test prompt. If the
request succeeds, DrowAI is using the NVIDIA-hosted GPT-OSS 20B endpoint.

## Troubleshooting

- If **Update** is disabled, confirm that the API key field contains the newly
  generated key.
- If the model is unavailable, confirm that the NVIDIA key is still valid and
  that the free endpoint has not reached its current usage limit.