A friend told me about a mobile homework app he was building out for his kids using the vibe-coding app, Cursor. In that convo, he mentioned that he wanted to add a speech-to-text component for them, so I told him I’d do some research and send him model recommendations. In less than 10 minutes, a clear winner emerged, and it was a free, open model: Canary Qwen. Let’s deconstruct how I arrived at that recommendation. Step 1: Fire Up My AI Strategy App Selecting the best model for a particular task requires getting really comfortable with AI model leaderboards. However, the problem with model leaderboards is they are all structured and formatted very differently. Most follow a predictable pattern of tables with sorted lists of models. However, few explain their metrics (aka benchmarks ) and those that do often bury them in an Arxiv paper or blog post (with, of course, no link to said paper or post).…