Our story

If you build on a model, you want one that won't change under you.

I'm Johnny Mobley. I started building systems for other people in 2000, when I was 16, and I've been doing it ever since. Along the way I automated most of my own work, and the tools I built for myself became these companies. Each one runs on rules I wrote down, not on paid AI calls, which is why it can cost so little.

Picking a foundation model for something that has to keep working comes down to a few plain questions: what does it cost per million tokens, how much can it read at once, can I run it myself, and is the version I tested going to stay the version I get. I built GreybeardAI to answer those without a sales call.

The comparison is a short list of vendor prices and context windows, each row linked to the vendor's own pricing page and dated, because those numbers move. The stability check reads a model's public Hugging Face history and hands you the exact revision to pin. It doesn't host any model and it isn't a guarantee.

What works today

• Filters 8 widely used models from Anthropic, OpenAI, Meta and Mistral by input price, context window, latency and whether you can self-host, with each row linked to the vendor's pricing page
• Checks how long a Hugging Face model has gone without a published change and gives you the exact revision to pin

What it doesn't do yet

Artificial Analysis still does things GreybeardAI doesn't:

• Benchmarks of model quality, speed and cost, measured independently.
• Hundreds of models and hosting providers. GreybeardAI compares 8.
• An API and data export.
• Industry reports and guides.

What it costs

Artificial Analysis's price is from their pricing page as of October 29, 2026. Prices change, so check theirs before you compare.

— Johnny