NVIDIA Build currently lists 138 models and shows 73 under its “Free Endpoint” filter. NVIDIA documentation describes these as serverless NIM API endpoints hosted on DGX Cloud, with OpenAI-compatible interfaces for large language model inference.
NVIDIA is offering free serverless API access to dozens of AI model endpoints through its NVIDIA Build catalog and NIM API documentation.
NVIDIA Build’s public models page currently lists 138 models in total and shows 73 models under a “Free Endpoint” filter. The catalog includes model families and entries such as GLM-5.2, Kimi-K2.6, Mistral, Gemma, Nemotron and DeepSeek V4, according to NVIDIA Build’s live model listing.
NVIDIA Build describes the service as “Free serverless APIs for development” on its Try NVIDIA NIM APIs page. That page also includes a “Get API Key” flow for developers who want to try supported models through NVIDIA NIM, NVIDIA’s system for serving AI models for inference.
NVIDIA’s API Catalog Quickstart Guide says the NIM endpoint is hosted on NVIDIA DGX Cloud and explains that users can obtain an API key from a model page. The guide also says the same process applies to any model in the NVIDIA API Catalog, indicating a shared access pattern across supported catalog models.
The safest reading of the public NVIDIA pages is that many, but not all, catalog models are available as free endpoints. NVIDIA Build’s models page lists 138 total models, while the “Free Endpoint” filter shows 73. That supports the claim that NVIDIA is offering free endpoint access to dozens of models, rather than to every model in the catalog.
For developers, a key practical detail is API compatibility. NVIDIA’s NIM for Large Language Models API Reference states that NIM exposes an OpenAI-compatible inference API. The documented interfaces include chat completions and model-listing endpoints.
That compatibility may reduce integration work for teams that already use OpenAI-style clients or application code. However, NVIDIA’s documentation does not imply that all models behave identically. Context windows, output limits, available parameters and response behavior can vary by model and by endpoint configuration.
NVIDIA Build’s model catalog spans several prominent model families. The cited catalog listing includes GLM, Kimi, Mistral, Gemma, Nemotron and DeepSeek entries among the available models. NVIDIA Build also provides filtering by endpoint type, including the “Free Endpoint” category.
Because the catalog is a live NVIDIA Build page, availability may change over time. Developers evaluating a specific model should check that model’s NVIDIA Build page for current endpoint status, limits and access details.
NVIDIA’s public pages position these endpoints as development resources. The Build page uses the phrase “Free serverless APIs for development,” and the NVIDIA Docs quickstart explains how to obtain a key and call hosted NIM endpoints.
The provided NVIDIA sources do not establish a blanket no-logging commitment for all requests sent to free endpoints. Developers working with confidential code, proprietary datasets or regulated information should review NVIDIA’s current terms, privacy notices and product-specific documentation before sending sensitive material to hosted endpoints.
The verified development is not that every model in NVIDIA’s catalog is free. It is that NVIDIA Build currently exposes a substantial set of free serverless model endpoints, and NVIDIA’s official documentation describes a common NIM API workflow with OpenAI-compatible large language model endpoints.
For developers, that combination makes NVIDIA Build a useful place to test hosted models without setting up local infrastructure. Based on NVIDIA Build and NVIDIA Docs, the current offer is best understood as free development access to dozens of cataloged models through documented NIM endpoints running on NVIDIA-hosted infrastructure.
Hero image prompt: Abstract editorial illustration of cloud-based AI model endpoints connected to developer workstations, with glowing server racks, neural network nodes and API connection lines in a green and dark graphite color palette; no logos, no real product screenshots, no readable text.
Hero image alt text: Abstract illustration of developers connecting to cloud-hosted AI model endpoints through API links.
NVIDIA is offering free serverless API access to dozens of AI model endpoints through its NVIDIA Build catalog and NIM API documentation.
NVIDIA Build’s public models page currently lists 138 models in total and shows 73 models under a “Free Endpoint” filter.
The catalog includes model families and entries such as GLM 5.2, Kimi K2.6, Mistral, Gemma, Nemotron and DeepSeek V4, according to NVIDIA Build’s live model listing.
Continue reading
