OpenAI has sent an official email announcing it will sunset the self-serve fine-tuning API for developers. Starting immediately, new users can no longer create fine-tuning jobs. Existing active users will have access only until January 6, 2027. As for already deployed fine-tuned models, their inference service will be tied to the lifecycle of the underlying base model — when the base model is retired, the service stops.
OpenAI’s stated reasoning: newer base models like GPT-5.5 are already extremely good at following instructions and formatting. Compared to expensive fine-tuning, directly writing prompts combined with retrieval-augmented generation (RAG) is cheaper, faster, and sufficient for the vast majority of use cases.
Tech media outlet Startup Fortune argues this isn't just product cleanup — OpenAI is pulling back on low-level model interfaces. Startups used to rely on cheap self-serve fine-tuning to build customized products for vertical industries. With this channel closed, developers are left with generic prompt engineering. That makes it much harder for startups to build proprietary technical moats around the OpenAI ecosystem. Meanwhile, OpenAI's Codex can now control your browser from behind the scenes — an indication the company is shifting focus to agentic features rather than customizable models.