OpenAI to Fully Close Fine-Tuning API: Large Models Completely Shift to Prompt, Startup Teams Lose Underlying Customization Path

According to Dynamic Beating monitoring, OpenAI has sent an official email announcing the complete shutdown of its developer-facing self-serve Fine-tuning service. Effective immediately, new users are no longer able to create fine-tuning tasks, and existing active users can only use the service until January 6, 2027. As for deployed fine-tuned models, their inference service will be tied to the lifecycle of the underlying base model, ceasing when the base model is retired.

OpenAI's core reason for this decision is that the new generation of base models (such as GPT-5.5) is already highly powerful in following instructions and formats. Compared to costly fine-tuning, directly writing prompts with Retrieval-Augmented Generation (RAG) is not only cheaper and faster but also sufficient for the majority of scenarios.

Technology media Startup Fortune believes that this is not merely a product cleanup but rather OpenAI is contracting access to its underlying models. In the past, startups often leveraged low-cost self-serve fine-tuning to build custom products for vertical industries. With this channel closing, developers can only use generic prompt engineering. This raises significant barriers for startup teams to establish customized technological moats around the OpenAI ecosystem.