With frontier and open-weight labs shipping new checkpoints every few weeks, I kept finding myself wondering two simple things: how many days ago did a given model line actually ship, and is a refresh due soon based on that lab’s usual pace?

Inspired by the MacRumors Buyer’s Guide, I put together ModelWatch. Instead of hardware buying advice, it tracks release history and expected refresh cadence across major AI model families from OpenAI, Anthropic, Google DeepMind, xAI, Meta, DeepSeek, Qwen, Mistral, Kimi, Z.AI, MiniMax, Microsoft, NVIDIA, and Cohere.

For each model track, it shows:

• Current age in days since the latest release
• Typical cycle using a rolling 5-interval median (so a single long historical gap or a quick 7-day point patch does not skew expectations)
• Prior releases and how many days each version stayed current, plotted on the same horizontal scale
• Direct links to the official release post, Artificial Analysis, and Arena leaderboards, with both English and Chinese (中文) views

All model data lives in a single schema-validated models.json file so adding a new release is a one-line CLI command and a git push.

• Live site: paul.mou.dev/modelwatch
• Source: github.com/moomou/modelwatch