OpenAI Shuts Down Developer Self-Service Fine-Tuning
OpenAI announced via official email that it will stop offering new self-service fine-tuning services for developers starting today, and new users will not be able to create fine-tuning tasks.
Existing active users can use the service until January 6, 2027. The inference service for deployed fine-tuned models will be tied to the lifecycle of the underlying base model, ceasing when the base model is retired.
OpenAI explained that new generation base models like GPT-5.5 are extremely powerful in instruction following and format control. Using Prompt+RAG is cheaper and faster than expensive fine-tuning and can cover the vast majority of scenarios.
Source: Public Information
ABAB AI Insight
OpenAI's move continues its strategy of shifting from "model openness" to "base control." Previously, it had gradually tightened developers' customization capabilities through API pricing and usage restrictions. Fine-tuning was a primary low-cost path for startups to create industry-specific models (such as in law, healthcare, and finance).
On the capital front, OpenAI is directing more developer traffic and payments towards general Prompt engineering and RAG solutions, reducing its own costs of training and maintaining fine-tuned models while strengthening control over the underlying base, forcing startups to rely more on the OpenAI ecosystem rather than building their own technological barriers.
Similar to how AWS gradually tightened certain underlying service interfaces and Meta moved from open Llama to a more weighted strategy, OpenAI is currently in the later stages of transitioning from a "model platform" to a "closed-loop AI operating system," significantly compressing the customization space for developers.
Structural judgment: This essentially represents a transfer of pricing power. By closing the self-service fine-tuning channel, OpenAI is reclaiming customization capabilities from developers, as the performance of the new generation base can replace most fine-tuning scenarios, concentrating capital, technological barriers, and long-term revenue towards OpenAI's core base and enterprise-level services, weakening the differentiation space for vertical startups.
ABAB News · Cognitive Law
Fine-tuning sells customization, Prompt+RAG sells scale control.
When the base is strong enough, open interfaces become optional.
Whoever controls the underlying model controls the ultimate pricing power of the entire ecosystem.