Last updated: October 13, 2026 · Sign-up flow, account verification, current error messages and free-tier limits re-checked against NVIDIA documentation, NVIDIA’s GitHub and the NVIDIA Developer Forums on October 13, 2026
Your NVIDIA API key — the same thing people call the NVIDIA NIM API key — is a free nvapi- token you create at build.nvidia.com: sign in with a free NVIDIA account (this enrols you in the Developer Program), complete the account verification if asked, open Settings → API Keys (or click “Get API Key” on any model) and click “Generate API Key”. One key works with the hosted NIM models in the catalog at no cost for prototyping. Usage is capped by model-specific rate limits rather than credits, and production use needs an NVIDIA AI Enterprise licence.
NVIDIA’s build.nvidia.com catalog gives developers free access to hosted AI models — Nemotron, DeepSeek, Llama, Qwen, embedding, vision and speech models among them — through an OpenAI-compatible API. You need no local GPU, only the key. This guide covers the October 2026 sign-up flow step by step, including the account and phone verification that now blocks many new users. It shows your first call in Python, curl and Node.js, lists the free limits as they stand today, and fixes the errors people actually hit: 403 Authorization failed, instant 429s and “verify your account” banners.
NVIDIA API key, NIM API key, nvapi key: are they the same?
Yes. “NVIDIA API key”, “NVIDIA NIM API key”, “NVIDIA AI API key” and “build.nvidia.com API key” all mean the same credential: the personal token that starts with nvapi- and is generated on build.nvidia.com. NVIDIA’s documentation describes it as the key that “authorizes HTTP calls to NVIDIA-hosted NIMs” at integrate.api.nvidia.com and ai.api.nvidia.com. You do not need a separate key per model: one key covers every hosted model your account is allowed to call. The only NVIDIA key that is genuinely different is the NGC personal key, which is used to pull self-hosted NIM containers. It is compared in its own section below.
If you are trying to understand what NIM itself is (hosted endpoints vs self-hosted containers, supported models, OpenAI compatibility), read our NVIDIA NIM API explained overview. This page is only about getting the key and making it work.
Is the NVIDIA API key really free?
Yes, for development and testing. NVIDIA made hosted NIM (NVIDIA Inference Microservices) endpoints free to NVIDIA Developer Program members, and enrolment costs nothing. You give a valid email, accept the terms and, on many accounts, verify a phone number. The key unlocks the hosted model catalog for prototyping, research and evaluation, and there is no credit balance to run down any more. What is not free is production: serving real end users or business transactions falls under NVIDIA’s production definition, which requires an NVIDIA AI Enterprise licence. The exact limits are in the free limits section below. For the full cost picture, see our NVIDIA NIM pricing & limits guide.
How to get a free NVIDIA NIM API key in 5 steps (October 2026)
The whole flow lives at build.nvidia.com and takes two to five minutes when verification goes through first time:
- Sign in or create a free NVIDIA account. On build.nvidia.com click Login, or click Get API Key on any model page. Enter your email and either sign in to an existing NVIDIA account or set a password for a new one. NVIDIA’s quickstart notes that signing up this way makes you a member of the NVIDIA Developer Program, which is free.
- Verify your account if build.nvidia.com asks. New accounts often see a verification prompt or banner before keys can be generated. Confirm your email, and if a phone verification step appears, enter your mobile number and the SMS code. Accept any API trial terms shown. If the SMS never arrives or you see “exceeded limits”, jump to troubleshooting. Retrying the same number over and over usually makes it worse.
- Open API Keys. Go to
build.nvidia.com/settings/api-keys(profile menu → API Keys), or stay on a model page and use its Get API Key button. Both open the same generator. - Generate and copy the key. Click Generate API Key and give it a recognisable name if the dialog asks (one key per project makes revoking easier). Then click Copy Key. The token starts with
nvapi-, and NVIDIA warns you may not be able to read the full secret again later, so save it now as an environment variable, not in code. - Make a test call. Point any OpenAI-compatible client at
https://integrate.api.nvidia.com/v1with the key and the exact model ID from a model page that offers a free hosted endpoint. Copy-paste examples are below.
NVIDIA API key vs NGC API key: which one do you need?
For calling hosted models on build.nvidia.com, you need the NVIDIA_API_KEY (nvapi-…) — not an NGC personal key. The two get confused because both come from an NVIDIA account and both unlock “NIM”, but NVIDIA’s NeMo Retriever documentation is explicit: the build.nvidia.com key “is not the same string as your NGC personal key used for Helm and nvcr.io access. Do not substitute one for the other unless your tooling explicitly documents that mapping.”
| NVIDIA API key (build.nvidia.com) | NGC personal key (org.ngc.nvidia.com) | |
|---|---|---|
| Where you create it | build.nvidia.com → Settings → API Keys (or Get API Key on any model page) | org.ngc.nvidia.com/setup/api-keys → Generate Personal Key |
| Typical env variable | NVIDIA_API_KEY |
NGC_API_KEY |
| What it authorises | HTTP calls to NVIDIA-hosted NIM endpoints such as integrate.api.nvidia.com and ai.api.nvidia.com |
Pulling NIM container images from nvcr.io, NGC Helm charts and, if you tick it, public API endpoints |
| Options at creation | Name; the secret is shown once | Key name, Expiration and Services Included (e.g. NGC Catalog and Public API Endpoints) |
| You need it when… | Prototyping against the hosted catalog from code, an IDE or an agent | Self-hosting a downloadable NIM on your own GPUs (Docker, Kubernetes, Helm) |
If you only want to call models from a script, stop at the build.nvidia.com key. If you later self-host a NIM container, generate a separate NGC personal key for the image pull — NVIDIA’s own RAG Blueprint docs select both NGC Catalog and Public API Endpoints under Services Included so one NGC key covers both jobs in that workflow. Keep the two keys in separate variables so a leaked prototyping key never grants registry access.
How to use your NVIDIA API key (first call)
NVIDIA’s hosted endpoint is OpenAI-compatible, so you can reuse the standard OpenAI SDK — just swap the base URL and the current model ID from the API Catalog.
First, store the key as an environment variable so it never lands in your source. Set NVIDIA_MODEL_ID to a currently listed catalog model before running the example:
export NVIDIA_API_KEY="nvapi-xxxxxxxxxxxxxxxxxxxxxxxx"
import os
from openai import OpenAI
client = OpenAI(
base_url="https://integrate.api.nvidia.com/v1",
api_key=os.environ["NVIDIA_API_KEY"],
max_retries=5, # the SDK backs off and retries 429 / 5xx responses
timeout=60,
)
resp = client.chat.completions.create(
model=os.environ["NVIDIA_MODEL_ID"],
messages=[{"role": "user", "content": "Say hello in one short sentence."}],
max_tokens=64,
)
print(resp.choices[0].message.content)
If you get a reply, the key works for that endpoint. From here you can swap the model string for another currently listed model and reuse existing OpenAI-based tooling. For picking a model and comparing against other providers, see our best inference APIs 2026 roundup.
No Python? The same call works from any HTTP client. With curl:
curl https://integrate.api.nvidia.com/v1/chat/completions \
-H "Authorization: Bearer $NVIDIA_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "'"$NVIDIA_MODEL_ID"'", "messages": [{"role": "user", "content": "Say hello in one short sentence."}], "max_tokens": 64}'
Many Node.js tools and proxies expect the key in a config.js file (people search for exactly “NVIDIA_API_KEY config.js”). Read it from the environment there too, so the file can be committed without the secret:
// config.js — never hard-code the nvapi- string here
module.exports = {
baseURL: "https://integrate.api.nvidia.com/v1",
apiKey: process.env.NVIDIA_API_KEY,
model: process.env.NVIDIA_MODEL_ID,
};
if (!module.exports.apiKey?.startsWith("nvapi-")) {
throw new Error("NVIDIA_API_KEY is missing or not an nvapi- key");
}
Which models can you call? The catalog mixes chat LLMs, embedding, reranking, vision and speech models from NVIDIA and third parties, and the list changes often — third-party roundups quote anything from “18” to “100+” free models depending on when and how they counted. Treat the model page on build.nvidia.com as the only authority: copy the exact model ID there, and if a page offers only a downloadable container, the hosted call will not work with your key.
Want to use the key inside an IDE or a terminal agent rather than a script? Our guide to using the free NVIDIA API in Cline, Continue, Aider, OpenCode and Claude Code has a ready configuration for each tool and a current list of NIM models that support tool calling.
NVIDIA API key free limits and credits (October 2026)
The free NVIDIA API key has no credit balance and no price tag. What caps it is a per-model rate limit that NVIDIA does not publish as one number. NVIDIA’s own forum staff confirmed the switch: “We no longer use a credit-based system for build.nvidia.com… this has been replaced by rate limits for trial usage.” Tutorials that still talk about starting credits or requesting more credits describe the old system.
| Question | Answer as of October 2026 | Where to check |
|---|---|---|
| What does the key cost? | $0. No credit card is part of NVIDIA’s documented quickstart flow | build.nvidia.com sign-up flow |
| Are there free credits? | No. Credits were replaced by rate limits for trial usage | NVIDIA Developer Forums (staff reply) |
| What is the rate limit? | Model-specific and not published as a universal figure. Users report their account page showing “Up to 40 RPM”, and popular models can return 429s below that | Your account / model page on build.nvidia.com |
| Can I raise the free limit? | No official way on the free tier (see below) | Paid, partner or self-hosted path |
| Self-hosted (downloadable) NIM | Free for research, development and testing on up to 16 GPUs through the Developer Program | NVIDIA NIM FAQ |
| Production use | Not covered by the free key; requires NVIDIA AI Enterprise (a 90-day evaluation exists) | NVIDIA AI Enterprise licensing |
In practice the free key is excellent for evaluating models, building demos and running a coding assistant at a human pace. It is the wrong layer for anything that needs guaranteed throughput: long autonomous agent runs, parallel workers, batch jobs or customer traffic. For AI Enterprise prices, the 90-day evaluation, 429 handling patterns and the point where self-hosting becomes cheaper, see the dedicated NVIDIA NIM pricing and rate limits guide. This page keeps only what you need to use the key.
Can you get a higher rate limit on the free tier?
Not through the forum, and not officially. Several guides repeat that you can “apply” to raise the free tier from 40 to 200 requests per minute, and the NVIDIA Developer Forums are full of such requests. The moderator answer quoted in those threads is consistent: “There is no official way to circumvent this rate limit or to receive a rate limit increase on that same tier. And specifically here on the forums we do not have any influence on those rate limits.” Higher throughput means leaving the free hosted tier for a self-hosted NIM, a partner endpoint or a paid NVIDIA path. If you just need more free headroom for prototyping, spreading work across providers is the realistic option. Our best free LLM APIs in 2026 comparison and NVIDIA NIM alternatives guide cover which ones publish their limits.
Anyone with your key can burn through your rate limit and act under your NVIDIA account. Never commit it to Git, never paste it into client-side JavaScript, and revoke it on build.nvidia.com if it leaks. Use environment variables or a secrets manager, and create one key per project.
Does the NVIDIA API key expire? Rotating and revoking keys
Plan for the key to stop working at some point, and make replacing it a one-line change. NVIDIA’s documentation for build.nvidia.com keys does not publish an expiry policy. Third-party guides disagree (“never”, 30 days, 12 months), so if the key dialog offers an expiration choice, the value you pick is the one that applies. NGC personal keys ask you to pick an Expiration at creation (NVIDIA’s RAG Blueprint docs choose Never Expire). For NGC keys NVIDIA documents two ways out of an expired or leaked key: create a new key and delete the old one, or use the Action menu to Rotate it. Either way, “you must update the new key information in your environment variables and code.”
- Generate the replacement first, in the same place you created the original key.
- Swap the value in your
.env, secrets manager or CI variable. This is why the key should never live in source code. - Test one call (the curl above) with the new key.
- Delete the old key. If it leaked to Git, revoking it matters more than rewriting history, because public repositories are scanned for secrets continuously.
NVIDIA API key not working? Fixes for current errors
Most “my NVIDIA API key doesn’t work” reports in 2026 are account problems, not code problems. The NVIDIA Developer Forums’ Access/Accounts board is dominated by three patterns: verification that never completes, 403 Authorization failed on chat completions, and 429 responses that arrive instantly. Find your symptom:
| Symptom | Likely cause | What to do |
|---|---|---|
| Banner “Please contact support to verify your account”, “No API Key Permissions” or “API Access Restricted by your Organization” when you try to generate a key | Account verification is incomplete, or your personal organization has not been granted API access yet | Finish email and phone verification, accept any API trial terms, then sign out and back in. If it persists, email help@build.nvidia.com (see the next row) |
| SMS code never arrives, or the phone number shows “exceeded limits” | Known problem with phone verification, reported mostly for non-US numbers (+86 and others); repeated attempts make it worse | Stop retrying the same number. Email help@build.nvidia.com with your registered email, every phone number you tried and a screenshot of the error. This is the route NVIDIA staff give on GitHub and the forums |
| 401 Unauthorized | Key missing, mistyped, wrapped in quotes or whitespace, revoked, or not loaded in this shell | Run echo ${NVIDIA_API_KEY:0:6}. It must print nvapi-. Re-export it, or generate a new key |
403 “Authorization failed” on /v1/chat/completions while /v1/models returns 200 |
Your account or personal organization lacks the hosted-inference (“Public API Endpoints”) permission. Less often, you are sending an NGC key created without that service | Confirm you are using a build.nvidia.com nvapi- key. Try a second model that offers a free endpoint. If every model returns 403, the block is on the account, so contact help@build.nvidia.com. Forum threads show no self-serve fix |
| 404 / model not found | Wrong model ID, a retired model, or a model that is downloadable-only (no hosted endpoint) | Copy the exact ID from the model page, or list IDs with GET /v1/models |
| 429 Too Many Requests, sometimes on the very first request | You hit the model’s rate limit, or a busy flagship model is throttling below your account’s nominal RPM | Back off and retry (the max_retries setting above), cap concurrency, and switch to a less congested model. The free limit is not raised on request |
| 5xx, timeouts or very slow first tokens | The hosted model is overloaded or degraded | Retry with backoff, lower max_tokens, and keep a fallback model ID in config |
A two-command check tells you whether a problem is the key or the account. If the first call fails, the key or its loading is wrong. If the first succeeds and the second returns 403, the key is fine and the account is missing hosted-inference permission:
# 1) Is the key valid at all? (expect HTTP 200 and a JSON model list)
curl -s -o /dev/null -w "%{http_code}\n" https://integrate.api.nvidia.com/v1/models \
-H "Authorization: Bearer $NVIDIA_API_KEY"
# 2) Can this account run inference? (expect 200; 403 = account permission, 429 = rate limit)
curl -s -o /dev/null -w "%{http_code}\n" https://integrate.api.nvidia.com/v1/chat/completions \
-H "Authorization: Bearer $NVIDIA_API_KEY" -H "Content-Type: application/json" \
-d '{"model": "'"$NVIDIA_MODEL_ID"'", "messages": [{"role": "user", "content": "ping"}], "max_tokens": 8}'
Choose your next NVIDIA NIM guide
Frequently asked questions
Is the NVIDIA API key free?
Yes. The NVIDIA API key from build.nvidia.com costs nothing and needs no credit card in NVIDIA’s documented flow. It is meant for development, testing and evaluation through the NVIDIA Developer Program. Production use, meaning real end users or business transactions, requires an NVIDIA AI Enterprise licence.
How do I get an NVIDIA API key?
Go to build.nvidia.com, sign in or create a free NVIDIA account (this enrols you in the Developer Program) and complete email and phone verification if asked. Then open Settings > API Keys at build.nvidia.com/settings/api-keys, or click Get API Key on any model page, and click Generate API Key. Copy the nvapi- token immediately, because you may not be able to view the full secret again.
Is the NVIDIA NIM API key the same as the NVIDIA API key?
Yes. NVIDIA API key, NVIDIA NIM API key, NVIDIA AI API key and build.nvidia.com API key all refer to the same nvapi- token. It authorizes calls to NVIDIA-hosted NIM endpoints such as integrate.api.nvidia.com, and one key covers every hosted model your account may call. Only the NGC personal key, used to pull self-hosted NIM containers, is a different credential.
How many free credits does NVIDIA give?
None any more. NVIDIA staff stated on the Developer Forums that build.nvidia.com no longer uses a credit-based system and that credits were replaced by rate limits for trial usage. Tutorials quoting a starting credit balance describe the old system.
What is the rate limit of the free NVIDIA API key?
NVIDIA does not publish one universal figure; limits are model-specific. Users report their account page showing up to 40 requests per minute, and busy models can return 429 errors below that. There is no official way to raise the limit on the free tier, so heavier workloads need a self-hosted NIM, a partner endpoint or a paid NVIDIA path.
Why does my NVIDIA API key return 403 Authorization failed?
If GET /v1/models works but POST /v1/chat/completions returns 403 Authorization failed, the key is valid but your account or personal organization lacks hosted-inference (Public API Endpoints) permission, often because verification is incomplete. Make sure you use a build.nvidia.com nvapi- key, try another free-endpoint model, and if every model fails, email help@build.nvidia.com.
Do I need a phone number or credit card for the NVIDIA API key?
No credit card is part of NVIDIA’s documented quickstart. Many new accounts, however, must verify a phone number by SMS before key generation is unlocked. If the code never arrives or the number shows exceeded limits, email help@build.nvidia.com with your registered email, the phone numbers you tried and a screenshot.
Is the NVIDIA API key the same as an NGC API key?
No. The build.nvidia.com key (NVIDIA_API_KEY, starting with nvapi-) authorizes calls to NVIDIA-hosted NIM endpoints. The NGC personal key (NGC_API_KEY) is created at org.ngc.nvidia.com and is used to pull NIM containers from nvcr.io and Helm charts. NVIDIA’s docs say not to substitute one for the other unless your tool documents it.
Can I use the free NVIDIA API key in production?
No. NVIDIA defines production as use beyond development, testing, research or evaluation, including real end users and business transactions. That requires an NVIDIA AI Enterprise licence. Use the free key to prototype and benchmark, then move to self-hosted NIM or a managed inference API with published quotas.
Does the NVIDIA API key expire?
NVIDIA does not publish an expiry policy for build.nvidia.com keys; if the key dialog offers an expiration option, the value you choose applies. NGC personal keys let you choose an expiration at creation. If a key expires or leaks, generate a new one, update your environment variables, test a call, then delete the old key.
Sources & methodology
Sources prioritise NVIDIA’s own documentation, catalog and developer forums. Current public guidance describes model-specific rate limits rather than a universal credit allowance; confirm access conditions in your own account. Forum threads are user reports unless the text says a staff member or moderator replied; we use them for current error patterns, not as NVIDIA policy. Links accessed September 6, 2026; sources 11–13 added September 29, 2026; sources 14–18 added and sources 4, 6 and 11 re-checked October 13, 2026.
- NVIDIA Technical Blog — Access to NVIDIA NIM Now Available Free to Developer Program Members. developer.nvidia.com blog
- NVIDIA — Try NVIDIA NIM APIs (model catalog). build.nvidia.com
- NVIDIA — API Keys settings. build.nvidia.com/settings/api-keys
- NVIDIA — API Catalog Quickstart Guide. docs.api.nvidia.com quickstart
- NVIDIA — NIM General FAQ (free vs production use). docs.api.nvidia.com product FAQ
- NVIDIA Developer Forums — API credits and model-specific rate limits for build.nvidia.com. forums.developer.nvidia.com
- NVIDIA — NVIDIA AI Enterprise (production licensing). nvidia.com AI Enterprise
- DecodeTheFuture — NVIDIA NIM API Explained (hub). decodethefuture.org
- DecodeTheFuture — NVIDIA NIM API Pricing & Limits. decodethefuture.org
- DecodeTheFuture — Best Inference APIs 2026. decodethefuture.org
- NVIDIA NeMo Retriever documentation — Authentication and API keys (NVIDIA_API_KEY vs NGC personal key). docs.nvidia.com
- NVIDIA RAG Blueprint — Get an API Key (NGC personal key options, expiration and rotation). docs.nvidia.com
- NVIDIA Developer Forums — Rate limit increase request: 40 RPM → 200 RPM (build.nvidia.com free tier), moderator response. forums.developer.nvidia.com
- NVIDIA Developer Forums — 403 Authorization failed on integrate.api.nvidia.com: /v1/models returns 200, chat completions 403, missing “Public API Endpoints” permission (Aug–Sep 2026). forums.developer.nvidia.com
- NVIDIA Developer Forums — “No API Key Permissions” and “verify your account” banner on a new account (Jul 2026). forums.developer.nvidia.com
- NVIDIA Developer Forums — 429 on every request below the “Up to 40 RPM” account limit (Jul 2026). forums.developer.nvidia.com
- NVIDIA Developer Forums — Resolving account access issues: email help@build.nvidia.com with registered email, phone numbers tried and screenshots. forums.developer.nvidia.com
- GitHub NVIDIA/NemoClaw issue #214 — API key generation blocked by phone verification for non-US numbers; NVIDIA maintainer response (Mar 2026). github.com

