Tokens or Towers

An independent guide to AI models, what they cost, and the hardware to run them. It tracks 267 models from 66 labs with 2,010 sourced scores, and plots capability against price, where a model that is good enough is often far cheaper than the leader.

The stack planner asks six questions about your workloads, data rules and budget, takes about 90 seconds, runs entirely in your browser, and returns three priced stacks that mix hosted models and owned hardware. Every number shows where it came from.

What is on it

• A leaderboard on a public intelligence index, with vendor-reported scores labeled
• A capability-for-price chart of the language models
• Team seat pricing for 2 to 500 developers
• A hardware grid: which models fit on which machines, and how fast they run
• A Method page on where the data comes from; the site calls itself a working prototype