If you are using the AI Assistant in Elements with an OpenAI or Anthropic API key, thought I would post the current pricing. You can get this pricing from OpenAI’s Docs page directly here and from Antrhopic’s Docs page directly here.
Personally I prefer using the MCP connection that comes built-in with Elements, as I have both an OpenAI and Anthropic subscription, and as of yet I haven’t burnt through my hourly or weekly limits when building websites in Elements.
For those that like using the AI Assistant directly in the app however, I can say that OpenAI’s Sol model is quite good for the more complex work in my experience, while the Luna model is great for the easier stuff like text generation, entering metadata, and even copying components and sections around.
It’s good to change the model depending on what task you are performing. You wouldn’t want to use the Sol model just to generate some metadata for your pages. Likewise you probably wouldn’t want to use the Luna model to build a custom component with a lot of complex functionality. I haven’t actually used the Astra model at all yet, as I haven’t built anything that would need that much level of intelligence. For most tasks I sort of feel that the Luna and Sol models are good enough. Of course depends on what you are building though.
Surprisingly, I haven’t really used Claude at all to build anything, as I’m finding OpenAI’s models do the jobs I need in the way I like. Probably a personal preference though, I might be Team OpenAI vs. Team Anthropic.
Anyway hope the above helps. I’m interested to hear what y’alls preferred AI model is when you’re building in Elements?
I’ve recently been using GPT 6.1 Sol on High (via MCP) to build custom components for the new Realmac site we’re building and it’s been doing an amazing job!
Check this out… I designed a grid and added text and images, then worked with Sol to produce a little custom component in Elements for each feature
I’m on a Pro plan for GPT, and never seem to hit the limits…
My problem is the marketing speak used in describing the models. No way to really parse what “focused, high-volume” or “complex work” or “most demanding work” really mean. What I think they really mean is “we’ll send you to our latest model and our fastest server” versus “we’ll just use our older model and our original, slower server.” Keeping track of exactly how models are trained and tuned has become near impossible. Sometimes you get hints (number of parameters, etc.), but it sure feels like choosing between 87 octane and 89 octane when the car’s manual says “use a higher-octane level.”
But I’m with @dan on this. Use MCP via chat on a Pro or Max plan if you want to be cost effective. I’m using Claude many hours every day and have never hit a limit. When I tried to do that via even their cheapest API, the token use quickly surpassed the monthly charge for chat. So, start with the lowest MCP plan and up it a level if you do run into constraints. But I’ll bet you never break US$100/month.
I’m still trying to figure out which open LLM (via LMStudio) works best with Elements. I keep coming back to QWEN, but every now and again it hiccups in ways I don’t find with Claude chat.
