What LogicMark actually is
LogicMark is a curated index of nineteen single-file prompts, grouped into five families: procedural SVG art, Three.js and WebGL scenes, playable browser games, self-contained web UI, and data visualisation. Each prompt is written so that a correct answer is one HTML or SVG file you can open directly, with no build step, no package install, and no external assets. That constraint is the point: it makes results comparable across models and easy to eyeball, because either the scene renders or it does not.
How a run works
Choosing a model equips it in the workspace exactly as if you had picked it in the Model Organizer, then hands the prompt to the chat surface with that model already active. There is no hidden evaluation harness and no scoring model sitting in between. You read the output yourself. Runs execute against whichever provider keys you have configured, so a run costs whatever that provider charges you and nothing more.
Why the prompts are shaped this way
Short prompts with a single deliverable expose different failure modes than long agentic tasks. A model that writes fluent prose about orbital mechanics may still emit a Three.js scene with the camera inside the geometry, an animation loop that never requests a frame, or an SVG whose viewBox clips the artwork. Those failures are visible in one glance, which makes this a fast first-pass filter before committing a model to a longer piece of work.
What it does not do
LogicMark does not currently record runs, score them, or rank models on a leaderboard. Published figures you see elsewhere for these models are vendor-reported and are not reproduced here as measurements. If you want a durable record of outputs and a side-by-side comparison between two models on the same prompt, that is a planned addition, not something the page does today.
Frequently asked questions
Is LogicMark a benchmark with scores?
No. It is a prompt launcher with a curated task list. It runs prompts and shows you the output; it does not grade the result or maintain a leaderboard.
Does running a prompt send my data anywhere?
Only to the model provider you selected, using your own API key. The prompt list and the page itself are static, and Accelerated Logic does not proxy or store your runs.
Which models can I run?
Any model equipped in your workspace, including catalog models, custom OpenAI or Anthropic keys, local Ollama endpoints, and in-browser WebLLM models.
Why are all the tasks single-file?
So the output is directly viewable and comparable. A single self-contained file removes build tooling and dependency resolution as confounding variables.