Non finito

Non finito is a specialized platform for directly comparing and evaluating the outputs of different AI models. Think of it as a public testing ground where you can see how models like OpenAI's o1, DeepSeek R1, or Grok 2 handle the same exact problem. You provide a prompt—which can be a text question, a math problem, or even an image with instructions—and Non finito runs it through multiple models, presenting their answers side-by-side for easy comparison. This allows you to see not just the final answer, but often the detailed reasoning steps each model took to get there.

What makes Non finito stand out is its focus on concrete, task-based evaluation. The site is populated with public 'evals'—specific test cases ranging from logical reasoning and entity tracking to visual question answering and base64 decoding. Each evaluation shows you which models passed or failed, and you can drill down to see the full, often lengthy, chain-of-thought. It turns abstract claims about model performance into something you can verify yourself. The interface is straightforward, presenting the model inputs and outputs in a clean, table-like format that highlights differences.

This tool is incredibly valuable for developers and researchers who need to choose the right model for a specific application. Before building a system that relies on AI for, say, interpreting price charts from images or solving complex math, you can use Non finito to test candidate models on similar tasks. It helps move beyond marketing specs to real-world performance, identifying strengths, weaknesses, and unexpected behaviors across the AI landscape. It's essentially a due diligence platform for the age of generative AI.