Using NanoGPT requires funds in your account. Add funds by going to your balance and depositing via credit card or using crypto.

To chat with ChatGPT, Claude or other models click Chat and type in the textbox at the bottom of the page. Our default model is ChatGPT-4o, you can change this at any time.

To generate an image or images, click New Image and type your prompt in the textbox. Our default model is Flux Schnell which gives 10 free generations per day currently, both the model and resolution can be changed.

No. We are not affiliated with Andrej Karpathy or his nanoGPT project. This NanoGPT service is an independent product.

Karpathy's nanoGPT is an open-source educational project that explores how to build and train your own GPT-style LLM (AI model).

As explained by ChatGPT itself, "a model is a type of AI that's trained to understand, interpret, generate, and respond to human language. It learns from a vast dataset of text and develops an understanding of language patterns, grammar, facts, and even nuances like humor or sarcasm."

We would describe these models as fantastic assistants on a wide range of topics: personal advice, programming, helping with learning, writing, medical questions and being an intelligent partner in practically any subject you can think of.

Some of the models we offer are open-source, some are closed-source. The famous ones such as ChatGPT and Claude are closed-source and are seen as the most advanced models.

Open-source models such as LLama and DeepSeek are generally cheaper, but do less well on most benchmarks.

Most models have different gradations. GPT-4 is "smarter" than GPT-3, while Claude 4 is smarter than Claude 3. Prices tend to reflect this.

Any text that is in the conversation so far gets sent to the model that you are using, every time. When you start a new conversation nothing except your new message gets sent. When you have a conversation where a few messages have gone back and forth, it means the previous messages are also sent to the model.

It does not matter for this whether you keep using the same model or switch halfway through. Switching between models halfway through the chat means model B gets sent your prompt along with the message history in that conversation so far with model A, and uses it all to generate a reply.

Yes, we offer prompt caching for Claude models. Enable it in settings by clicking "Adjust Settings" or the gear icon.

Prompt caching stores frequently used content (system prompts, conversation history) in memory, providing faster responses and up to 90% cost savings on subsequent requests. It's ideal for document analysis, RAG applications, and long conversations where you repeatedly reference the same content.

Claude's strength at analyzing long documents makes prompt caching particularly valuable for research and knowledge work involving existing documents or structured data.

For both video and image models, you can adjust various settings by clicking the "Adjust Settings" button (gear icon on mobile) when you're in image or video generation mode. This allows you to customize parameters like resolution, aspect ratio, and other model-specific options to better suit your needs.

Alternatively, visit our Media Generation page which is better suited to image and video generation.

To generate videos, select a video model from the model dropdown when creating content. You can choose from various video generation models that are ranked based on quality and performance. Simply enter your prompt describing the video you want to create, and the AI will generate it for you.

Alternatively, visit our Media Generation page which is better suited to image and video generation.

The supported video formats and lengths depend on the specific video model you choose. Different models may offer various resolution options, durations, and output formats. You can adjust these settings using the "Adjust Settings" button when in video generation mode.

Alternatively, visit our Media Generation page which is better suited to image and video generation.

If you're using NanoGPT without creating an account, you can still backup your important data:

For your funds: Go to Settings to find your UUID (unique identifier). You can use this UUID to restore your balance if you ever lose access to your device or clear your browser data.

For your conversations: Export your chats from the Conversations page and save them locally. You can later import these conversations back into NanoGPT from the same page. Note that conversations are stored in your browser's localStorage (accessible via F12 → Application → IndexedDB → database → conversations). We do not store conversations.

We strongly value privacy. You are not required to create an account or provide any personal information to use this website.

By default we do not store anything except a cookie containing your account number with your associated balance. Chats you have via this website are stored locally in your browser. When you clear your website data, we cannot help you retrieve your conversations or images. We do not store them, we do not keep them as a backup.

When we communicate with model providers we send your prompt and the prompts you've sent previously in that conversation. No conversation is linked to any other conversation, and no identifiable information is sent along with the prompts. From the provider's perspective each conversation is a standalone conversation that is only identifiable by our API key, which is the same for all users and all conversations.

Across all providers we use, we apply a minimum-retention and maximum-deletion posture by using the strictest retention and deletion settings they offer. Providers still run their own infrastructure and legal/compliance policies, so their provider-side terms ultimately govern what they retain.

For more information on how we deal with privacy see our Privacy page in which we explain it more thoroughly.

Our web search uses LinkUp by default. LinkUp publishes a privacy-focused no-log policy for search queries. You can switch to Tavily or Exa in Settings if you prefer a different provider.

Web search is a separate service from our AI models and operates independently from other privacy features like TEE (Trusted Execution Environments). Provider-specific privacy policies apply to the web search portion.

When you use web search with a TEE (Trusted Execution Environment) model, it's important to understand that these are two separate services with different privacy characteristics.

TEE models provide stronger data-in-use isolation and attestation for model execution. However, the web search portion is handled by the provider you select in Settings (LinkUp by default, with Tavily and Exa available), which is separate from the TEE inference path.

Web search queries are not processed within the TEE environment. When you combine TEE models with web search, TEE protections apply to model inference and do not automatically extend to web-search requests.

Yes! If you spend $500 or more on text models in a single day, you automatically qualify for a 10% discount on text models the following day. This discount is recalculated daily.

The discount applies to all text/chat models (like GPT-4, Claude, Gemini, etc.) and is applied automatically. You can see if you have this discount active in your Settings page under "Discounts".

Note: This discount does not stack with referral or subscription discounts, but does stack with the 5% Nano payment discount.

ChatGPT, Claude and other AI assistants assist many people with everything from everyday tasks to complex solutions. They have been proven to be incredibly helpful tools in furthering knowledge, teaching new skills, and broadening perspectives.

Not everyone has access to these models. With global median income at ~$250 per month, a $20 per month subscription is expensive. Billions of people are unable or unwilling to pay this much, limiting their ability to use state-of-the-art models.

Many people do not have access to a credit card. Some might live in countries where you don't want anyone to know what questions you are asking ChatGPT.

This service is for those people. Rather than paying upfront, people pay only for what they use. NanoGPT can be used anonymously, by anyone, from anywhere. We accept credit cards but also offer the option to add via crypto for those without credit card access.

ChatGPT, Claude and other models should be available to anyone in the world and we are trying to make that happen.

Nano offers instant payments at zero fees. It's an environmentally friendly cryptocurrency focused on scaling payments. It's possible to try Nano for free by downloading a wallet like Nautilus (iOS, Android) and getting some free Nano from a Nanodrop faucet.

This Substack article provides further reading on Nano while the Nano Foundation website offers more content.

We use Nano because it offers instant transactions at zero fees to anyone in the world. It's very easy to develop on, with broad documentation and quick testing since all transactions are feeless and provide immediate feedback.

Nano's vision is to be a digital currency for everyone. We strongly believe in the vision of allowing anyone in the world to transact instantly without fees, without giving away personal details and without requiring anyone's permission.

We also see Nano as a store of value, and would rather hold a hard form of money with a fixed supply than a form of fiat money.