which-llm
Artificial Analysis has a ton of data on language models. which-llm makes that data available to an agent, so I can ask about the models, compare them, and make charts like this one.
It combines the benchmarks with OpenRouter prices, context windows, supported inputs, and the model names you actually pass to the API. You can ask for a fast model that can read images, compare a few specific models, or see how much intelligence you're giving up to spend less. The agent runs Python commands to get the relevant data.
This is intelligence against the cost per task on that same benchmark. Up and to the left means a higher score for less money. The frontier connects the models where you can't improve one of those without giving up the other. You can also plot speed, context length, or token prices, depending on what you care about. Benchmark cost tells you what happened on the benchmark; your own project's cost depends on how you use the model.
The data refreshes daily, and the installed skill downloads a new snapshot when its copy gets old. A fresh copy works offline, with no API key needed. I recently expanded the scraper to include more of Artificial Analysis: individual evaluations, coding agents, and image, video, and speech leaderboards. There was a lot more there than we were making available through the skill.