Live Data

The Adoption Ledger

What the market runs, and when it dies

Everyone measures which models are popular. Nobody measures how long they have left. We take live gateway adoption share and cross it with our own deprecation calendar — because the model with the biggest share of production traffic can still have a shutdown date, and that gap is where projects break.

Running on borrowed time

Models with real adoption and a scheduled shutdown. This is the join nobody else publishes: popularity from the gateway, mortality from the calendar. If something here is in your codebase, it's a migration you haven't scoped yet.

Loading the ledger…
⏱️

The trap is the pin, not the model. Adoption is measured against a model's public name; shutdowns are announced against dated snapshot IDs like gpt-5-nano-2025-08-07. Call the rolling alias and you migrate automatically. Pin the dated ID — as most production code does, for good reproducibility reasons — and you get cut off on the date instead. wreck-watch scans your repo for exactly that, and the deprecation clause makes it billable when it happens.

What's actually being used

Share of gateway traffic on the latest day. Three different questions: requests is who serves the most calls, tokens is who does the most work, spend is where the money goes. They disagree, and the disagreement is the interesting part.

Loading…

Rising and fading

Share change across the window. This is the churn a consultant actually feels: the model you standardised on last quarter quietly sliding down the table.

Loading…

This is public JSON

/adoption/data.json — rankings, movers, and the at-risk join, refreshed daily. Cite it, chart it, build on it.

Or query it from your agent: the Project Wrecked MCP server already exposes the deprecation half. Put the calendar in your CI with wreck-watch.

What this table means for a delivery, and why the risk is in the pinned ID rather than the model: Your most-used model has a death date.

Free, and staying free. If your company gets commercial value from this dataset, sponsoring it keeps it running — and buys no influence whatsoever over what it says.