On 23 October, OpenAI switched off a generation of older models. On 14 January 2027, GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini go too. I collected 16,130 AI repositories created on GitHub between 1 and 24 October, downloaded the code of the 12,838 that were available and read every model name in it. In 2,377 of them, a model is called by name in code written that month.
Of the 721 whose default model is OpenAI's, 65.5% default to one already switched off or with a shutdown date (statuses as of 26 October). The most common default, in nearly one in four, is gpt-4o-mini.
At the median, that OpenAI default was 18.2 months old on the day the repository was created. For Anthropic it was 1.5 months, for Google 4.2.
Coding agents know their own family. In code co-signed by Claude that calls Anthropic, 6.8% default to a dead or dying model. In Claude-signed code that calls OpenAI, 67.5%, about the same as code with no agent signature (64.5%).
What I took from it as a builder: the model name is the one line that stops working without any change to your code. Keep it in one config file, check it in the build against the provider's own list of models (GET /v1/models), and add one line to your coding agent's instructions file saying which models are current.
Full method and data in the X Article: https://x.com/con_golias/status/2114750216809140276
