NVIDIA’s Free API Move Is Bigger Than It Looks. Here’s Why

NVIDIA’s Free API Move Is Bigger Than It Looks. Here’s Why

Someone asked in the comments after my last piece: is there a way to use GLM-5.2 without any quota at all?

I thought that was a big ask. GLM-5.2 is a serious model. Running it costs money. There’s no such thing as truly unlimited access to something like that.

Then I looked into it.

Turns out there is. And the provider isn’t a Chinese startup. It’s NVIDIA. The GPU company. And they have a free trial API portal — 77 models including GLM-5.2, available right now.

My first reaction was disbelief. Still is, honestly.

Let me just sit with that for a second. NVIDIA. The company that makes H100s at $30,000 a pop. Running a free trial API portal for AI models. That’s not what I expected to find when I went looking for GLM access.

Here’s what I found.

Build.nvidia.com hosts 141 AI models. Seventy-seven come with free trial API endpoints — not paid, not metered in the usual sense. Real endpoints you can hit to evaluate the models.

The list is worth seeing to believe. DeepSeek-V4-Pro. Kimi-K2.6. MiniMax-M2.7. Qwen3.5. And GLM-5.2, listed as Free Endpoint plus Downloadable, provided by Z.ai — the company behind it. Two million downloads in the first 24 hours, according to the platform.

Most people skip past this platform. They hear NVIDIA and think GPUs, gaming rigs. They don’t think free AI API playground.

But it exists.

I know what you’re thinking. “I’m not a developer.” Stick with me. There’s a no-code path at the end.

Here’s what I wrote about before — and why this is different.

CatPaw gives you 500 calls before you fill out a form and wait for approval. QoderWork gives you 2,000 credits that expire in 30 days. Based on the comments, people burn through those credits in a few days of real use. These aren’t scams. They’re real free tiers. But they’re capped.

The NVIDIA trial is different. Here’s how it works.

Go to build.nvidia.com. Register with Google or GitHub. Verify your phone number — it accepts mainland China numbers, +86, SMS works fine. I’ve done it. Takes seconds.

Once verified, go to your profile, find API Keys, hit Generate. Give it a name. Copy the key — it starts with nvapi- and only shows once. Close the page and it’s gone.

The page says it’s a trial. No daily limits mentioned in the interface, but “trial” is the word NVIDIA uses. I’ll be straight with you: I don’t know exactly what the rate limits are after prolonged use. I haven’t hit a wall in a week of casual testing. But this is NVIDIA’s infrastructure — I wouldn’t assume it’s truly unlimited under heavy production load.

Here’s what nobody is asking: why is NVIDIA doing this?

Everyone writes about how to get the key. Nobody asks why a US hardware company is hosting free trial endpoints for Chinese AI models.

My take: it’s a platform play.

NVIDIA’s real business is GPUs. But the people using their GPUs need software to run. If NVIDIA becomes the place where developers discover and try models — any model, from anywhere — then when those developers need compute, NVIDIA is the first call.

Kind of like how Google gives Android away for free. Not out of generosity. Because controlling the developer surface matters more than the short-term cost.

That’s what I think is happening here. I could be wrong.

Three things to get started. The base URL is https://integrate.api.nvidia.com/v1. The model name is on the page — just copy it as shown. Your API key, which you already have.

Put those three into Claude Code instead of Sonnet. GLM-5.2 handles the work. No subscription. No per-call charges for the trial tier. Takes about ten minutes to set up if you’re following along.

There’s a catch. More than one.

Z.ai provides the model. NVIDIA hosts it. When something breaks, you’re probably dealing with Z.ai. The platform is NVIDIA. The model is someone else’s. That line matters when things go wrong.

“Trial” is also a specific word. I’ve used it casually for a week without hitting a wall. But I haven’t run production workloads. If you need reliability guarantees, this is a free trial, not a contract.

The bigger question I can’t answer yet: how long does this last? Trials change. Terms shift. The key I’m using today might behave differently in six months.

That’s the reality of free infrastructure with no SLA.

If you’re a developer who wants to experiment with GLM-5.2 without watching a counter, this is the most practical free path I’ve found. It’s not obvious. It involves an API key and a URL. It’s not a polished app.

But it works.

If that sounds like too much — there are tools like CC Switch that handle the setup with a graphical interface. I plan to cover that separately.

The person who asked about quota-free access was right to ask. The answer was just hiding in the developer portal of a GPU company, of all places.

That’s the part I keep coming back to.

NVIDIA. The last place you’d expect to find a trial key to a Chinese AI model.

Whether it stays free is the question I don’t have an answer to yet.