OpenAI Enacts New Self-Serve Fine-Tuning Restrictions Effective July 2, 2026

OpenAI will restrict self-serve fine-tuning access starting July 2, 2026, requiring organizations to have run inference on a fine-tuned model within the past 60 days to create new jobs. This follows a May 7, 2026 rule barring new fine-tuning for organizations without prior history. The phased deprecation, ending January 6, 2027, aims to manage resource allocation and encourage active use of fine-tuned models.

OpenAI Enacts New Self-Serve Fine-Tuning Restrictions Effective July 2, 2026 OpenAI has implemented the next phase of changes to its self-serve fine-tuning platform. Effective July 2, 2026, organizations that have not run inference on a fine-tuned model in the past 60 days will no longer be able to create new fine-tuning jobs. The update follows the May 7, 2026 notification that first limited access. On that date, creating fine-tuning jobs or training became unavailable to organizations that had never previously run fine-tuning. The July 2 restriction tightens eligibility further by requiring recent inference activity on fine-tuned models. Key Timeline from OpenAI Deprecations Page May 7, 2026 : Creating fine-tuning jobs or training unavailable to organizations without prior fine-tuning history. July 2, 2026 : Creating fine-tuning jobs no longer available to organizations without inference on a fine-tuned model in the past 60 days. January 6, 2027 : Active existing customers lose the ability to create new fine-tuning jobs. Inference on fine-tuned models remains available until the underlying base model is deprecated. Inference on existing fine-tuned models continues without interruption until the specific base models themselves reach end-of-life. OpenAI states that impacted customers receive email notifications and documentation updates for these changes. The restrictions apply specifically to OpenAI’s self-serve fine-tuning offering. Organizations relying on custom models for proprietary workflows, specialized performance, or competitive differentiation now face stricter ongoing activity requirements to maintain creation access. High-intent users—particularly AI-forward businesses and technical teams—should verify their recent inference activity against the 60-day window ahead of the July 2 cutoff. Existing fine-tuned models and their inference capabilities remain supported through base model deprecation timelines. This phased approach gives advance notice while gradually narrowing self-serve fine-tuning availability. Full details and the complete deprecation schedule are available on OpenAI’s official API Deprecations page.