NVIDIA Is Giving Away 120+ AI Models for Free for a Year — I Tested It and It's Insane

NVIDIA has opened up 120+ AI models on build.nvidia.com as free APIs for a full year — no credit card required. OpenAI-compatible endpoints running on DGX Cloud (H100/B200 clusters). I tested DeepSeek V4 Flash, Nemotron, MiniMax, Kimi, and everything works. The catch: 40 RPM rate limit, which is plenty for normal use.

💡 What You Will Learn

NVIDIA has opened up 120+ AI models on build.nvidia.com as free APIs for a full year — no credit card required. OpenAI-compatible endpoints running on DGX Cloud (H100/B200 clusters). I tested DeepSeek

NVIDIA Is Giving Away 120+ AI Models for Free for a Year — I Tested It and It's Insane

NVIDIA recently made a massive move — opening over 120 AI models as free APIs on build.nvidia.com, unlimited usage for a full year, no credit card required.

Yes, the same NVIDIA that can not keep GPUs in stock is offering their DGX Cloud clusters — H100s and B200s — for developers to use for free for a year.

I spent an afternoon registering, configuring, and thoroughly testing it. Verdict: it is genuinely amazing, and surprisingly so.

What is the deal? Why is NVIDIA being so generous?

This is about NVIDIA NIM (NVIDIA Inference Microservices). NIM is NVIDIA's infrastructure for packaging AI models as microservices, running on their own DGX Cloud.

Previously, to use these models you had to either buy your own hardware or pay per-call through third-party APIs. Now NVIDIA is opening them directly: free, unlimited, no enterprise verification. Individual developers can get an API key with just an email + phone verification.

The platform now hosts 140+ models, of which 77+ are labeled 'Free Endpoint' — directly callable at no cost. The API is OpenAI-compatible, meaning virtually any AI tool on the market can connect to it.

Notable models worth using

DeepSeek V4 Flash — 284B MoE, 1M token context, excellent at coding. Previously required registering and topping up at DeepSeek's own portal. Now free on NVIDIA's platform.

Nemotron Series (NVIDIA's own) — Nemotron-3-Ultra-550B is NVIDIA's flagship, hybrid Mamba-Transformer architecture, 1M context, exceptional at agent reasoning and tool calling.

MiniMax M2.7 — 230B parameters, strong at coding, reasoning, and office tasks.

Kimi K2.5 — Moonshot AI's star model, top-tier Chinese long-text understanding.

Qwen Series — Qwen-Image (text-to-image) and Qwen-Image-Edit (image editing) both have free endpoints.

Others — GLM-5 (Zhipu), Llama (Meta), Mistral series... pretty much every well-known open model has a free endpoint here.

Testing: step-by-step setup

Step 1: Register. Go to build.nvidia.com, click Login. Enter email, verify. Then bind a phone number. Gotcha: Chinese +86 numbers sometimes show 'limit exceeded' — email help@build.nvidia.com for manual verification, usually resolved within hours.

Step 2: Get API key. Avatar to Settings to API Keys to Generate Key. Copy and save it.

Step 3: Configure. OpenAI-compatible protocol, so setup is trivial. Base URL: https://integrate.api.nvidia.com/v1

For Hermes Agent: set model.base_url and model.api_key, then select a model name.

I tested DeepSeek V4 Flash on Hermes — response speed was surprisingly fast. DGX Cloud H100 clusters deliver lower latency than many paid APIs.

Is 40 RPM enough? The free tier limits to 40 requests per minute. In practice, this is plenty for daily use — coding, analysis, research. 40 RPM means one request every 1.5 seconds, which is faster than a normal human can type queries.

The free models cover code generation, multimodal understanding, image generation, embeddings, RAG, tool calling, and voice. For most individual developers and small teams, this eliminates the biggest barrier to entry for AI-powered development.