Currently, the main type of model I use is Large Language Models. I encountered this newfangled technology not-too-long-ago and found it very intriguing indeed. I have followed the news since December 2025, and have had my hand at using various models, including local, cloud and open-source models.
I use a mix of OpenAI's ChatGPT (including Codex), Anthropic's Claude (including Code), Google's Gemini (including the Google AI Studio API), Alibaba's Qwen (both locally and on the web), and am experimenting with GLM, Grok, Deepseek, MiniMax and Mistral.
I have come up with my own measures of testing a model's capabilities (sort of a benchmark of sorts), and would like to have my own database of quality, cost and efficiency for use with my Gauss project.
I try to make it clear where I use Artificial Intelligence (including the Harmonica project, and the Gauss project). However I also wish to clarify that I have not used it at all in the creation of this website. I also only ever send personal emails and messages without any AI written text, wholely written by myself.
I don't use Image Models much, hence this section is shorter, but I want to state one thing, which is that currently my benchmark for when an image model becomes acceptable is when it can draw the Lockheed CL-1201 flying in formation over London. With accurate visuals (to the design documents we can see), and also accurate geography over London.
So far, as much as an AI lab has tried to advertise their image model to me (OpenAI and Google in particular have been quite insistent), they have all failed (it is either a flying rectangle, or some futuristic thing nothing like it). I appreciate that this is an area with little to no reference images, but that is why I have chosen it as my benchmark. As I find it unlikely that this could be achieved, I believe that it will be quite some time before image models can live up to their advertisement.