I understand that you rely on many different providers for models, including open-source models in the subscription. However, it isn't clearly written anywhere which models are hosted by which providers. (Or maybe I'm just not looking hard enough.) Any chance you could disclose this information somewhere on the site? Like, "Primary provider for X is Fireworks, fallback Chutes", as an example. Adding a note about it for the "Available Models" list under the "</> API" tab would feel intuitive, for example.

I noticed your site and API are behind the Cloudflare proxy/CDN. Cloudflare inspects and routes all visitor traffic, which can expose users’ IP addresses and request data to a third party and centralizes metadata that could be retained or accessed. For better privacy and to avoid third‑party interception of user requests, please consider removing the Cloudflare proxy (or configure DNS only / self‑hosted CDN) so visitors connect directly to your origin servers. Thank you.

I remember that NanoGPT used to advertise "zero data retention" agreements with their LLM providers. At some point this language has been removed (or I'm misremembering) and replaced with "minimum data retention", which is a far more nebulous term, though even "zero data retention" does not have a standardized definition among providers (especially regarding abuse detection). Can you disclose which providers you have special data retention agreements with and what those terms are? It would go a long way to help the transparency of this platform.

Is there a way to see supported TPS for each model like OpenRouter displays?

Please consider adding generation settings such as Min-P, this is really helpful since most of prompts and presets for roleplay comes along with generation settings and its bad that nanoGPT API does not support Min-P setting.

More filter options for models (all of them) would be useful, such as "can be coaxed easily into NSFW", "does not guarantee privacy", etc...

I'm primarily using PayPal, so having it as a payment option would be nice

Sometimes a reasoning model gets interrupted, and does not finish the task. Currently i make them continue by sending a message that just says "continue". Maybe it's better to provide a separate button to continue generating.

Qwen3 Embedding 0.6B A powerful and cheap Embedding model that is also open source and available here: https://huggingface.co/Qwen/Qwen3-Embedding-0.6B, it would also be nice to obtain vectors without spending API requests.

As open source models are becoming an alternative for commercial models, especially for coding, the reliability needs are also increasing. The subscription provides a good way to have access to such models. For normal chat completion requests, "cheap" providers (your backend providers) are usually sufficient. But for coding, high reliability is needed (reasoning, tool calling). Some providers apply settings to the models that make them nearly impossible to use for coding. Therefore I suggest a pro or coding subscription with the same / similar limits like the normal subscription, but with providers who don't heavily modify the model settings (usually the model dev providers).

You guys are a little too fast at adding new features to the website/API. I often only find very useful features (like this new suggestion board) months after they launch because they don't seem to be announced anywhere other than your Discord. Can you please merge what you announce on Discord and the updates section on the website into a unified changelog? Ideally this would be backdated to include everything you've added thus far as well.

Allow finer control of API key usage, including daily RPD and daily NANO consumption allowed.

Can you contact the OpenCode team so we can use all our subscription models, as they aren't currently listed.

Hello! Z-Image Turbo is super cheap to run and is good enough quality for most of my needs, where the other models in the subscription don't quite meet my requirements. It would be super awesome if that model could be included in the subscription please! Thank you

I think it would be a good addition to include details about when a model was last updated or uploaded. Why? For example, the recent appearance of the Mistral Large 3 model. It appears in the available models list, but I'm not sure if it's the new version or an older one. While it might have "2512" at the end of the model name, this doesn't guarantee a specific date, as the other existing models don't have anything like that.

Models like GLM 4.7 and MiniMax M2/M2.1 now recommend passing `reasoning_content` or `reasoning_details` back into the context for enhanced reasoning performance. See https://docs.z.ai/guides/capabilities/thinking-mode and https://platform.minimax.io/docs/guides/text-m2-function-call. Is this possible with NanoGPT's OpenAI-compatible API? Or does it perhaps need an Anthropic-compatible API? And if so, can I suggest some clarity in the docs on how this might work?

Would it be possible to add embedding models to the subscription?

For those of us less organized, it would be nice to have option in all conversations tab to select multiple chats from a given page and have the option to apply a function from the three dot menu that you would normally only be able to apply for a chat one at a time and not even be able to select another without returning back to all conversations screen.

It would be great to add all Qwen3 embedding and reranking models to the subscription: Qwen3-Embedding-8B Qwen3-Embedding-4B Qwen3-Embedding-0.6B Qwen3-Reranker-8B Qwen3-Reranker-4B Qwen3-Reranker-0.6B Currently, such models are becoming increasingly necessary for codebase indexing and RAG tasks. As far as I have seen, competitors already offer these models within their subscriptions.

Would be nice to add Google drive files straight to the model instead of downloading first

Many providers do offer prompt caching nowadays and utilizing that yields great discounts. Unfortunately, through NanoGPT, we can't benefit from those (except with Claude, which is the most inconvenient to implement). That remains a big advantage of OpenRouter. For me personally, the most interesting providers are Moonshot AI, Google (Gemini) and DeepSeek. I am aware there are some suggestions like this already, but they can't be voted on, so this is my way of expressing demand. I'm also aware this might hurt your profits, if you only pay discounted rates while you charge base costs, but that would also mean the pricing isn't fair.

Either on a per-conversation level (where the web-search/thinking/memory toggles are), a per-preset level, globally, or through prompt tag eg <CURRENT_TIME> (in the manner you already have <CURRENT_DATE>) Plz I beg, I need my character to stop hallucinating timestamps and arguing when I say it's X time but they think it's Y. And it's a big boost for immersion in general when prompt engineering ❤️

Maybe people will pay for it, monthly limits can be toned down for those two models.

Or gift cards/voucher codes that can be purchased with cash from somewhere.

It would be great to have suno available for music generation in nano-gpt. I love your work!

Hello NanoGPT team! A suggestion I have is to provide vouchers like Mullvad does so people can deposit money anonymously without needing crypto. I didn't want to deal with the tax headache of cryptocurrencies, so I used the stripe process to add funds. I was under the impression that personal information would not be knowable by NanoGPT, but I can see my personal information under billing so you surely must have access, right? Vouchers would enable a clean separation for those like myself which do not want to rely on trust.

Your application already supports over 500 LLMs, and the Custom CivitAI update got me thinking - it would be incredible if users could access HuggingFace models through NanoGPT. Because RAM and GPU prices are rising, an AI rig is impossible for the average consumer, myself included. Adding this feature would be perfect because users could access more models and use them without getting expensive hardware. The basic idea would be to find a HF URL, like https://huggingface.co/mradermacher/Huihui-GLM-4.7-Flash-abliterated-i1-GGUF (a currently inaccessible model) that contains GGUFs and then to paste and submit it. Like Custom CivitAI, the LLM would have to be resolved to be used, but once finished you could start using it. Likely all users would enjoy this feature.

It would be cool to have a badge in the model list for open-source / open-weight models, similar to the (old?) badges for vision-enabled models, or maybe another way to highlight them (colored model names perhaps), as the description for models can’t really be seen on a mobile device. Why? Because I like to run open models most of the time and consider them as having better privacy, but still want to be able to use proprietary models should I want to. It’s a bit of a pain to know whether they are open source or not. This might clutter the UI, so I guess it can also be an opt-in/opt-out setting?

There are understandable privacy concerns with relying on Google/Gemini behind Polychat for Memory. I believe there should be an option to use the regular open source model providers, which have far better privacy practices.

Gemini natively supports grounding/google search api and is arguably more powerful than third-party search tools such as the supported tavily or linkup tools. Currently, it cannot be used through the nano-gpt api Supporting the native grounding api would allow for better search performance additionally, supporting more tools like google's code execution tool would allow for more use cases, matching the official api's capabilities

Please add an option to protect past conversations with pincode or 2FA

Currently there are no rejected submissions and the placeholder text says "Be the first to submit an idea!", implying to suggest an idea to get rejected. Would be better to replace the text with something like "All of your ideas are good so far!". Or reject this idea. That would be funny.

Due to the free, subscription-based GLM providers being way too slow sometimes, I'd like it if the thinking/reasoning version of the Original GLM 4.7 was also offered. I *could* just get it directly from z.ai, but... I'd rather get all my API's from here, neat and organized.

Sup.Ai is doing something really cool that I bet you guys could implement. They give the user the ability to submit to multiple models, which you have, but instead of just returning the results, they pass all three results through a model to evaluate the responses and construct a consensus using the best aspects of each response. Then, they return the final consensus to the user. The idea is that while one model might hallucinate or write poorly, it's unlikely that they all would at the same time. I would love to, for example, be able to combine Deepseek, GLM and Kimi K2 or something like that. The biggest trick, here, would be how you prompt the final model to generate a useful consensus, but if you pulled it off, it would be incredible. Another advantage is that if a model is unavailable or unwilling to complete the task, another model will, so the user never experiences a dropped response.

I would love to see Ministral 8b and Ministral 3b added, as well as the Mixtral 8x7b & 8x22b Models, if that would be possible

I want to make the API pricing easier to understand. This includes showing the price for image inputs and for video generation with or without audio. Some models do not have pricing, so I also want to clearly show which features each model can access, such as whether image input is allowed.

As a user, I'd like to see more detailed information for each log in the Usage page, including provider, latency, throughput of each log/usage.

Provide speed information and reliability information next to each provider.

Some of the subscription models take forever to start replying. Would be nice to know how fast they are in advance.

Staging work is often needed for working with reasoning models as it is impractical to sit and wait for the reply to be completed. Often however, it is easy to forget you were expecting the reply while you do other things, wasting time ultimately that could be used to have the LLM do more work or iterate. It would be quite useful to have a (very gentle, not annoying) little chime sound as an option that plays immediately after message completion. It would be nice to have several options for the sound so one that is pleasing can be chosen.

The chat page on mobile is far too crowded, particularly with the top bar, down. It's hard to read and navigate because the bottom bar for text input is so tall.

AI isn't aware by default of the current date, time and timezone of the user. For providing time-sensitive suggestions, an option to auto-add the users current time to prompts would be helpful. Then the AI could respond: for this interaction in this business communication, get back with the client on X of Jan at ~14:00 As an extension to this: Said response from the AI could be used to sort the conversations list, so that those contacts that require interaction earlier show above the ones with later dates. A more general approach to dynamic titles would be for the user to instruct the AI to respond in a certain format -like some sort of front-matter - that holds the next date to follow up. NanoGPT could interpret hints like this (and others) and adapt it's views accordingly. This are effectively three suggestions that could be considered separately. Feel free to split as you see fit.

Automatically choose the aspect ratio for generated image based on the original image.

I believe you said you're going to bring this back, that was a while ago though

Perplexity as Search Option (BYOK)

Please consider adding 2FA using authenticator apps. This would act as an additional layer of security.

Problem: some client libraries for OpenAI compatible APIs expect format specification for embeddings to work (base64 is an option officially). Example for such a library: OpenAI official nuget package

Hi, I would be very useful, if you would allow to upload multiple images as references for seedream as the model allows for it. I need to generate multiple reference images elsewhere but would prefer nanogpt

I read situation about those 3-5% users who caused abuse of service and unlimited tokens, and understand introduction of weekly limits fo 30mil tokens. But even among those 95% people can be few % of loyal users who may need extra tokens per week (especially when using agents or reasoning models). I would like to suggest a feature to introduce weekly limit size depending on how long user has been subscribed. It may be flat bonus after X months (i.e. 60m after 3-6 months) or incremental (i.e. +3-5mil with each month). Kinda a reward system for loyalty. And abusers usually will create a lot of new accounts and probably would not patient enough to cause issues.

Another random model request (sorry!) but would you consider adding Christian Bible Expert v2.0 24B? Should be on Featherless AI. It should be small and cheap and would help me a lot 🙏 https://huggingface.co/sleepdeprived3/Christian-Bible-Expert-v2.0-24B