OpenAI has notified developers that it will shut down its fine-tuning platform, blocking new training runs immediately and disabling the feature for all customers starting January 6, 2027. Inference on existing custom models will continue until their base versions are fully deprecated. The company is shifting focus from weight modification to RAG, prompt engineering, and orchestration, consolidating logic within the OpenAI ecosystem to simplify updates and billing. Developers relying on fine-tuning for niche adaptations will need to rethink their architecture.

Source