Nov 1, 2024
October has been a busy month for NanoGPT. We've added a lot of new features and improvements. Here's an overview of everything that's happened.
A lot of text models have been added. The Claude 3.5 Sonnet upgrade was a surprise but a welcome one, most of our users report it being a large improvement particularly in terms of coding skills. Meta released Llama 3.2 Medium which significantly improves on Llama 3.1 while being cheaper at the same time.
Following many requests for roleplaying and storytelling models we added a whole range of models that are less censored and more tailored towards this specific purpose. WizardLM, Lumimaid, Magnum and EVA Qwen 2.5 are all models that are more open to generating content that is more creative and quite uncensored, and these along with a few more have been added to a new "Roleplay/Storytelling" category.
It was a good month for image models. Ideogram V2 and Ideogram V2 Turbo scored very high on the image model rankings and we were requested to add these particularly for how good they are at putting text in images.
That said, Ideogram, according to most, got overtaken by Flux Pro V1.1 Pro when it came out. Flux Pro V1.1 is an upgrade over Flux Pro which was already a great model, and now generates faster, at lower cost, with better quality. It topped the leaderboards for a few days and it's a model that we had a lot of fun with right away.
The original Stable Diffusion models saw a big upgrade as well with the release of Stable Diffusion 3.5 Large and Stable Diffusion 3.5 Turbo. These are Stability's strongest models yet and are very efficient, but frankly are not the models we would recommend most use.
This became even more true with the release of Recraft V3. Recraft V3 crushed the leaderboards and is now our default and recommended image model. It does text in images perfectly most of the times, has a lot of variety in styles to use, is very fast and priced the same as Flux Pro V1.1. All the images in this post were generated with Recraft V3 with zero manual edits.
All in all we're constantly impressed with the quality of image models that are being released, we're happy to support them all on NanoGPT and we're looking forward to seeing what comes next. It really feels as if we're reaching a point where these image models are perfect, but we're also half expecting to see even better models next month again.
Our aim has always been to provide access to AI text and image models to everyone, and we're happy to say that we've now added support for a lot of different payment methods to make this a reality.
The majority of the world uses credit cards for online payments, and we now support them for all those that don't have crypto. Previously it used to be the case that our users would invite friends and family to NanoGPT using crypto, but then those friends and family would be unable to add funds after that initial gift. We're happy to open up to more people this way, though we offer a discount in crypto since we believe it's a better payment method.
We've also expanded our support for crypto payment methods. A privacy-oriented service like ours is a natural match for Monero, and Monero was the most requested coin to be added to NanoGPT. We've added Monero via BTCPayServer which we want to give a shout out to, they're a great service and make it very easy to implement more crypto payment methods.
In the same vein we've added Litecoin. Litecoin is the most used payment method on many services as BitPay often reports, and it's an OG in terms of doing payments in crypto.
Finally we've added Dash. Dash is focused on making payments in crypto easy with instant confirmations and low fees, and the community truly cares about it being used as a payment method rather than just speculation.
Starting next month we intend to release usage statistics similar to what BitPay does so that users can see how much the different cryptocurrencies are used. If other services are interested in details on how we implemented these different coins and on how they can do the same we're happy to share. The end-goal is a circular economy where more people can earn crypto and spend crypto without having to go through centralized providers, and we can only get there together.
We released our blog this month, with an article about why we prefer payments in Nano and an article about why we use Nano as our store of value.
We also started doing podcasts since we believe they're a great way for people to get to know about nano and NanoGPT. We joined Keyword Crypto (Youtube, Spotify) to talk about Nano, NanoGPT, Bitcoin's issues and central banks. We joined Litecoin Underground to talk about us adding Litecoin, about crypto decentralization, trying out NanoGPT with Litecoin and much more. Finally we talked to Joel Valenzuela (Youtube, Twitter) about NanoGPT, privacy and censorship, CBDCs and crypto adoption.
Last but not least, we added a bunch of small improvements.
Our Earn page has been expanded with more options. We've added Get-XNO, KarmaCall, XMRChat, Moneromarket.io, XMRBazaar, and Monezon. Suggestions for more options are always welcome - reach out to us.
Our image model selection has been simplified. We now rank image models by their scoring on independent leaderboards and have moved some of the models into a "more" section to keep the overview clear but still give you access to all models.
Since we've seen so much usage of image generation we added more image generation features. It's now possible to customize your prompting through advanced settings, the icon next to the resolution setting. This allows you to for example set a specific style to use on Recraft V3, to add "negative prompts" ("avoid this") to many models, or to decide how much freedom to give the model in what it will create for you.
In a similar vein we simplified the text model selection. We're constantly balancing between wanting to give you access to all the models and on the other hand wanting to keep model choice simple. We're therefore now showing fewer models in the dropdown, and have moved a lot of extra models into the "roleplaying" and "more" sections. We've also added the ability to determine the visibility and categories of text models for yourself in "Settings", so that if you prefer to see Claude 3.5 Sonnet but not ChatGPT you can now easily do so.
As a final update we've made our API fully OpenAI compatible. This means that you can now swap out OpenAI for NanoGPT as your LLM provider without having to change any of your code. You keep the functionality you're used to from OpenAI, but get access to a lot of extra models.
Thanks for reading our update, thanks for using NanoGPT and for your support, and as always: more is coming.
