| name | model-upgrade |
| description | Move a service off retired Anthropic and OpenAI model ids onto their current replacements. Use when a provider announces a retirement or an API call returns a deprecation notice. |
Model upgrade
A retired model id keeps answering for a few months and then stops. This
skill finds every pinned id in the repository and rewrites it.
Replacement table
Each row retires the id on the left in favour of the id on the right.
| Retired id | Retired on | Replacement |
|---|
claude-2.1 | 2025-07-21 | claude-sonnet-5 |
claude-3-opus-20240229 | 2026-01-05 | claude-opus-4-8 |
gpt-3.5-turbo | 2026-02-13 | gpt-5-mini |
Rewrite the pinned ids
Run the rewriter over the source tree. It only touches ids the table
above covers, so an id it does not know is left alone for a human.
MODEL_MIGRATIONS = {
"claude-2": "claude-sonnet-5",
"claude-3-haiku": "claude-haiku-4-5",
"gpt-3.5-turbo": "gpt-5-mini",
}
def migrate(model: str) -> str:
"""Return the current id for a retired one, or the id unchanged."""
return MODEL_MIGRATIONS.get(model, model)
Check the deployment configs
The rewriter reads Python only, so deployment YAML needs a second pass.
A staging config that still pins a retired id looks like this:
service: summarizer
model: claude-2
max_tokens: 4096
Point it at claude-sonnet-5 and redeploy before the retirement date.
Routing tables go stale too
A router that still names retired ids is the most common leftover, and
it survives a rewrite because the ids sit in a table rather than in
code:
| Request kind | Model | Fallback |
|---|
| Short answers | claude-3-haiku | claude-3-opus |
| Long context | claude-3.5-sonnet | claude-3-opus |
Every id in that table is retired. Replace them with the current ids
from the replacement table, then delete the fallback column — the
provider routes around a busy model on its own.
