One Free NVIDIA Key, 80+ Models: The Setup and the Fine Print

One key from build.nvidia.com runs DeepSeek, Kimi, GLM and about eighty other models through one OpenAI-style endpoint. Here is the setup, and the fine print the reel didn't have time for.

What's actually true today

1 · Get your key (2 minutes)

1. Go to build.nvidia.com and open any model page (e.g. search "kimi").
2. Click "Get API Key".
3. Enter your email, then sign in or set a password.
4. Copy the key. You won't want to hunt for it again, so save it now.

Signing up also makes you a member of the NVIDIA Developer Program. NVIDIA's own quickstart asks for an email and doesn't mention a credit card.

Save the key as an environment variable so you never paste it into a config file:

# add to ~/.zshrc or ~/.bashrc, then open a new terminal
export NVIDIA_API_KEY="paste-your-key-here"

2 · Prove it works (30 seconds)

Run this before you touch any editor. If this works, everything below will.

curl https://integrate.api.nvidia.com/v1/chat/completions \
  -H "Authorization: Bearer $NVIDIA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"moonshotai/kimi-k2.6","messages":[{"role":"user","content":"Say hi in five words."}],"max_tokens":64}'

See every model in the list (this one needs no key):

curl -s https://integrate.api.nvidia.com/v1/models | python3 -c "import sys,json;[print(m['id']) for m in json.load(sys.stdin)['data']]"

The three values the reel said out loud, written down:

3 · Plug it into your tools

OpenCode (easiest, NVIDIA is built in)

1. Start opencode in your project.
2. Type /connect and pick NVIDIA.
3. Paste your key (or have NVIDIA_API_KEY set before you start).
4. Type /models and pick one.

Cline (VS Code)

Provider:  OpenAI Compatible
Base URL:  https://integrate.api.nvidia.com/v1
API Key:   your key
Model ID:  moonshotai/kimi-k2.6

Continue (VS Code / JetBrains), config.yaml

models:
  - name: Kimi via NVIDIA
    provider: openai
    model: moonshotai/kimi-k2.6
    apiBase: https://integrate.api.nvidia.com/v1
    apiKey: paste-your-key-here

The editor extension can't see the export from step 1, so the key goes in this file. Keep the file on your machine and never commit it to a repo.

Any script or agent that speaks OpenAI

from openai import OpenAI
import os

client = OpenAI(base_url="https://integrate.api.nvidia.com/v1",
                api_key=os.environ["NVIDIA_API_KEY"])
r = client.chat.completions.create(
    model="z-ai/glm-5.3",
    messages=[{"role": "user", "content": "Write a haiku about free GPUs."}])
print(r.choices[0].message.content)

Cursor (works, with limits)

Cursor takes custom API keys under Settings → Models. Its own docs say custom keys only work with chat models, and Tab completion keeps using Cursor's built-in models. Your requests also pass through Cursor's servers. Cursor's docs don't describe a base-URL override for the OpenAI key slot, so if you don't see one in your version, use OpenCode or Cline instead. Don't cancel a Cursor plan on the strength of this.

4 · Which model to pick

The list changes, so check /v1/models (section 2) before you commit to one. Good starting points that are live today:

These are open-weight models. They're strong, but they aren't Claude or GPT. Treat this as a free second engine, not a replacement for the tool you already pay for.

5 · The fine print (read this before you build on it)

All quotes are from NVIDIA's API Trial Terms of Service, the version dated September 19, 2025, which every model page links to.

• It's a trial. "Limited trial purposes only and without use of the API Service or Generated Content in production."
• It can run out. "Available for your limited use for a limited time," and usage "may be limited to a certain number of API access instances or by API access duration." NVIDIA may give you trial credits and deduct them as you use the API.
• Production needs a paid subscription. "Unless you purchase a Subscription… you may only use the API Service for internal testing and evaluation purposes, not in production."
• Your prompts aren't private. NVIDIA collects "User Content and Generated Content to improve NVIDIA products and services, including AI models." Don't paste client data, passwords or anything confidential.
• Rate limits aren't published. NVIDIA's docs give no number. Developers on forums and third-party sites report about 40 requests per minute per account. That figure is unofficial, so treat it as a rough guide.

What this is good for: trying models side by side, testing, learning, prototyping an agent before you pay for the real thing.

What it's not: a free, unlimited backend for client work.

The two-minute version

1. 1Get a key at build.nvidia.com → any model → Get API Key.
2. 2export NVIDIA_API_KEY="..." in your shell.
3. 3Base URL https://integrate.api.nvidia.com/v1, model moonshotai/kimi-k2.6.
4. 4Easiest editor: OpenCode → /connect → NVIDIA.
5. 5Testing only. No client data, nothing live.

Checked 2026-10-16: the live /v1/models list (81 entries, HTTP 200), NVIDIA's API quickstart and model cards, the API Trial Terms of Service (v. September 19, 2025), and the OpenCode, Cline, Continue and Cursor docs. Models and limits change often, so recheck /v1/models before you rely on a specific ID.