NanoGPT

What is NanoGPT?

NanoGPT is a single OpenAI-compatible API in front of a large model catalogue: 634 text models are listed on its pricing page as of 31 August 2026, alongside image, video, audio, transcription and embedding models. Some open-weight models are self-hosted; closed-source models from OpenAI, Anthropic and xAI are passed through. It also ships a web interface, so the same balance works for chat and for API calls.

Pricing

Text model usage is billed at each model's list price with no percentage added on top, and there are no deposit fees. The operator gives GPT-5.5 as a worked example: $5/$30 per 1M tokens at OpenAI, and $5/$30 here. Every chat completion and Responses API response returns the exact cost charged, so the claim is checkable per request. Two exceptions are opt-in and documented: manually pinning a specific provider adds 5%, and bring-your-own-key requests are billed at 5% of the normal model cost. The no-markup statement is made for text specifically; image, video and audio models are priced per model on the same page without an equivalent claim.

An optional $12/month subscription includes 60M input tokens per week on most open-source models, usable through both the web interface and the API. The operator points at GLM, Xiaomi Mimo, Minimax and DeepSeek as the models it suits. Pay-per-prompt needs no subscription; balances start at $0.10 topped up in crypto or $1 by card.

Anonymous use and data handling

An account is not required. Without one, funds are topped up and spent directly, and conversation history is stored locally in the browser via IndexedDB rather than server-side. A permanent key restores access to the same balance from another browser or device. Registering stores only what the user consents to, and deleting an account deletes the data tied to it.

Getting started

The API is OpenAI-compatible, so an existing integration moves over by changing the endpoint URL and the key. A batch API, a benchmarks page and a public model directory are all available without signing in.

Who it is for

Developers who want breadth of catalogue behind one integration, and who would rather pay list price per model than hold a subscription with each provider. The trade-off is the usual aggregator one: routing sits between the caller and the model provider, so provider-specific features and support run through NanoGPT rather than the upstream vendor. The catalogue also includes uncensored and NSFW models, which is a deliberate part of the positioning and may matter either way depending on the use case.

Most used

The operator reports Claude Opus 5, Claude Fable 5 and Kimi K3 as the three most-used models on the platform (August 2026).