Web console for NVIDIA’s free API (build.nvidia.com). Chat, model comparison arena, streaming, and embeddings. No credit card, no queue, no catch.
# Switching from OpenAI to NVIDIA = 2 lines from openai import OpenAI client = OpenAI( base_url="https://integrate.api.nvidia.com/v1", api_key="nvapi-XXXXX" # free at build.nvidia.com ) r = client.chat.completions.create( model="meta/llama-3.1-70b-instruct", messages=[{"role": "user", "content": "Ola!"}] ) print(r.choices[0].message.content)
NVIDIA makes 100+ AI models available via a free API at build.nvidia.com. This project is a complete web console for using these models without any hassle.
Uses the library openai default (Python/JS). Switch base_url e api_key and you’re all set. No code changes.
Compare up to 8 models side by side with the same prompt. See each response’s TTFT, total time, and quality in real time.
Create an NVIDIA Developer account and generate the key nvapi-... and start using it. 40 requests/minute per model, with no charge.
The console comes with 8 models ready to use. The NVIDIA API has 100+ in total, and you can add any of them to the registry.
| Model | Provider | Type | Highlight |
|---|---|---|---|
| Llama 3.1 70B | Meta | LLM | Versatile, 128k context |
| Llama 3.1 405B | Meta | LLM | Largest Llama, top quality |
| DeepSeek R1 | DeepSeek | LLM | Reasoning, 671B MoE |
| Qwen 3.5 397B | Qwen | LLM | 17B active MoE, multilingual |
| GLM-5.2 | Z.ai | LLM | 128k, long response |
| Nemotron 120B | NVIDIA | LLM | 1M context, tool use |
| Mixtral 8x22B | Mistral | LLM | European MoE, fast |
| Llama 3.2 90B Vision | Meta | Vision | Analyzes images + text |
Embeddings also available via endpoint /api/embeddings (model nvidia/nv-embedqa-e5-v5).
The API is 100% compatible with OpenAI. The console is a polished frontend that connects to it via FastAPI (local) or Vercel (production).
Ongoing conversation with a selected model. SSE streaming shows tokens in real time. Message history is kept in the frontend.
Select 2-8 models, send them the same prompt, and compare their responses side by side. TTFT and total time for each model. Perfect for deciding which model to use.
Few dependencies. Everything is free.
Create a free account and generate an API key.
# Go to: https://build.nvidia.com/settings/api-keys # Click "Generate API Key" # Save the nvapi-... key
Install dependencies with pip.
python3 --version pip install openai fastapi uvicorn python-dotenv
To clone the project and deploy it.
git clone https://github.com/inematds/nvidiaapifree cd nvidiaapifree
Follow the steps below. By the end, you’ll have the console running locally with 8 free AI models.
Go to build.nvidia.com, create an account (no credit card required), and generate an API key.
# 1. Open in your browser: https://build.nvidia.com/settings/api-keys # 2. Click "Generate API Key" # 3. Copy the key (format: nvapi-XXXX...) # The key appears only ONCE. Save it immediately.
Clone the repo and create the file .env with your key.
git clone https://github.com/inematds/nvidiaapifree cd nvidiaapifree cp .env.example .env # Edit the .env and paste your key: echo "NVIDIA_API_KEY=nvapi-SUA_CHAVE_AQUI" > .env
The project uses FastAPI + OpenAI SDK. One line does it.
pip install openai fastapi uvicorn python-dotenv
Run the test script to confirm the key works.
python3 test_api.py # Expected output: "ok, working"
Start the FastAPI server and open the console in your browser.
python3 server.py # Server running at http://localhost:8812 # Open in your browser: http://localhost:8812
In the console, choose a model and start chatting. Or switch to Arena to compare models.
# Chat: select a model from the dropdown, type, and send # Arena: click the "arena" tab, select 2+ models, and send the same prompt # Models respond in parallel with real-time streaming
The API also works directly in your Python/JS code.
# Python from openai import OpenAI client = OpenAI( base_url="https://integrate.api.nvidia.com/v1", api_key="nvapi-SUA_CHAVE" ) # Chat r = client.chat.completions.create( model="meta/llama-3.1-70b-instruct", messages=[{"role": "user", "content": "Explique IA em 2 frases"}], temperature=0.7, max_tokens=512 ) print(r.choices[0].message.content) # Embeddings emb = client.embeddings.create( model="nvidia/nv-embedqa-e5-v5", input=["texto para vetorizar"], extra_body={"input_type": "query"} ) print(f"Dimensao: {len(emb.data[0].embedding)}")
The project is ready to deploy on Vercel. It works on the free plan.
Connect your GitHub repo to Vercel and import it.
# In Vercel: # 1. New Project > Import Git Repository # 2. Select inematds/nvidiaapifree # 3. Framework: Other
Add NVIDIA_API_KEY in the project's Environment Variables in Vercel.
# Under Project Settings > Environment Variables: NVIDIA_API_KEY=nvapi-SUA_CHAVE_AQUI # Optional: separate keys by model NVIDIA_API_KEY_LLAMA=nvapi-... NVIDIA_API_KEY_DEEPSEEK=nvapi-...
Everything you need to know about the NVIDIA NIM API in one table.
# Base URL https://integrate.api.nvidia.com/v1 # Chat Completions POST /v1/chat/completions # Embeddings POST /v1/embeddings # List models GET /v1/models
# Rate limit 40 RPM por modelo # scalable to 200 # Cost $0 # no credits, no charges # Credit card Nao requerido # Key expiration Nao expira
What’s already done and what’s next.