Your model name is not a promise

The string you pass as model feels like a fixed thing. It is not. Two changes this week show how a model can move under an app that never shipped a line of code.

DeepSeek swapped the model behind old names. On October 17 DeepSeek released DeepSeek-V4.1-Flash-1017. Its pricing page now says to use deepseek-flash as the model name, and that the legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted, but the models behind them have been retired. Requests to those names are now served by the new model and billed at the Flash price. If your app calls one of them, it got a different model today without an error.

OpenAI is turning off a batch of snapshots and aliases on October 23. OpenAI's deprecations page lists models that shut down that day, including gpt-4-turbo, gpt-4.1-nano, gpt-image-1, o1-2024-12-17, o3-mini-2025-01-31 and o4-mini-2025-04-16, along with short aliases such as o1, o3-mini and o4-mini, and several older fine-tuned models. On October 14 it also gave notice that four GPT-4o and GPT-4.1 snapshots, and fine-tunes built on them, leave the API on January 14, 2027.

What to do about it:

• Pin a dated snapshot when the provider offers one. A dated name like gpt-4.1-2025-04-14 tells you exactly which model you tested. An alias can be repointed or retired.
• Watch shutdown dates in code, not in your inbox. OpenAI's deprecations page says the List models endpoint (GET /v1/models) returns an optional shutdown_date field for each model. A daily check that fails loudly when a model you use gets a date is a few lines of code.
• Treat a provider-side change as a model change. Keep a small set of real prompts with expected outputs and rerun it whenever a name you use is repointed, the way DeepSeek's legacy names were this week.
• Check where your evals live. OpenAI's same page says existing evals on its Evals platform become read-only on October 31, 2026, and the dashboard and API are scheduled to shut down on November 30, 2026. If your regression tests live there, export them now.
• Stay off preview models in production. OpenAI says models with preview in the name may be retired with much shorter notice, such as two weeks.

The takeaway: the model name in your config is a pointer the provider controls. Pin what you can, monitor what you can't, and keep a test set you can rerun in minutes.

MODEL did not run code against either API for this piece. Everything above comes from DeepSeek's and OpenAI's own documentation, read on October 17.