GitHub's code search finds about 341,000 files that contain the string claude-sonnet-4-5-20250929. On the Claude API, that model retires on November 30, 2026, 18 days from today. Ten days ago the same search said about 287,000.
So the number of files naming it went up by about 54,000 while its end date got closer. That is roughly 5,400 a day.
The counts
| Exact string | Nov 2, 2026 | Nov 12, 2026 |
|---|---|---|
"claude-sonnet-4-5-20250929" |
about 287,000 | about 341,000 |
same, without Bedrock's anthropic.claude-sonnet-4-5-20250929
|
about 311,000 | |
image: postgres:14 |
about 72,000 | about 89,500 (Nov 11) |
Every number is GitHub's own count on its code search page, read on the date in the column. You can rerun the first one yourself:
https://github.com/search?q=%22claude-sonnet-4-5-20250929%22&type=code
Why split out Bedrock
The same model has more than one end date. Anthropic retires claude-sonnet-4-5-20250929 on the Claude API on November 30, 2026 (Anthropic's model deprecations). Amazon Bedrock calls it anthropic.claude-sonnet-4-5-20250929-v1:0 and runs it until April 8, 2027 (Bedrock model lifecycle). A file with the Bedrock ID does not break on the 30th, so it should not be counted as if it does.
The search "claude-sonnet-4-5-20250929" NOT "anthropic.claude-sonnet-4-5-20250929" leaves those out. About 311,000 files remain. Each of those names the ID that the Claude API stops answering on November 30. Some will be docs, changelogs and forks. Many will not.
What the count can and cannot tell you
The honest objection: GitHub's numbers are approximate. They move when the index changes, they include forks and vendored copies, and "files" is not "projects". A rise of 54,000 could be partly the index catching up.
I take that seriously, so here is what I think holds. The direction is the same for a model and for a database image, two very different things, measured the same way. And PostgreSQL 14 had its retirement date published five years in advance. Its count still rose by about 17,500 files in the nine days before it ended.
A published end date does not stop new code from naming the thing. Nothing in a repo reads that date. Not the code, not the linter, and not a coding assistant whose training data predates the announcement.
The next one
On December 11, 2026, OpenAI's API shuts down gpt-5, gpt-5-mini, gpt-5-nano and o3 (OpenAI deprecations). The search for "gpt-5-2025-08-07" OR "gpt-5-mini-2025-08-07" OR "gpt-5-nano-2025-08-07" returns about 268,000 files today. Azure calls the same IDs, and on Microsoft Foundry gpt-5-2025-08-07 runs until February 9, 2027, so that count carries two dates, not one.
If you want to know whether you are in the 311,000:
git grep -n "claude-sonnet-4-5-20250929"
The replacement Anthropic names is claude-sonnet-5-5.
I'll count again on December 1, the morning after.
Have you ever found a retired model ID in your own code after it stopped answering? Where was it hiding?
I'm Nabil. I build Since, which tracks what your code depends on and tells you before it breaks.
Top comments (0)