Reasoning demos
Each one opens up a different facet of how modern reasoning models work — their strategies, their failure modes, and the scaffolding that shapes both. Open any to run it live.
Three open-source models compete to guess your secret word. Swap their prompt scaffolding to see how harness quality shapes reasoning.
Same prompt, two agents. One rambles in markdown; the other can only call tools. See how a structured output schema makes hallucinations unrepresentable.
Watch a vision-capable model parse messy handwritten forms into structured data — and see where it second-guesses itself.
Multiple LLM personas — designer, critic, integrator — collaborate on a problem. Reasoning is shown step by step.
An LLM routes a sprint of coding tickets across a churning fleet of worker models, learning which fits which from outcomes alone — and rewriting its own routing rules each night.