Model lifecycle · Docs

Open-weight models move fast. We add new releases quickly and retire old ones on a fixed, predictable schedule, so you can plan upgrades instead of reacting to them.

Available. The model is listed in the catalog and in kurrens.ai/models.json. New releases are announced in the Changelog.
Deprecated. We announce the retirement date at least 30 days in advance: by email to accounts that used the model in the previous 30 days, in the Changelog, and on kurrens.ai/models/retirements. The model keeps working normally until then. Its entry in models.json gains a deprecation_date, which OpenRouter also reads.
Retired. From the retirement date, requests to the model return 404 with the error code model_retired and the recommended replacement.

What you’ll get in the notice

The retirement date (at 13:00 UTC).
The recommended replacement model and anything that differs: context window, precision, capabilities, price.
For dedicated endpoints, a migration window agreed with you — dedicated models are not retired on the public schedule.

Handling retirement in code

{
  "error": {
    "type": "invalid_request_error",
    "code": "model_retired",
    "message": "deepseek-ai/DeepSeek-V4-Flash was retired on 2026-11-30. Use deepseek-ai/DeepSeek-V4-Flash-0731."
  }
}

Catch 404 responses with code: "model_retired" and switch to the replacement, or pin model ids in configuration so an upgrade is a one-line change.