Leader in artificial intelligence OpenAI has released a feature that lets developers and corporate clients fine-tune GPT-4o, the company’s most sophisticated model, using their data for $25 starting on September 24, 2024.
The corporation declared that the new functionality is now fully functional in a blog post. The fine-tuned custom version, available for GPT-4o, is intended to let companies and developers customize the model’s replies to match certain requirements, like changing the structure and tone or adhering to intricate domain-specific guidelines.
OpenAI’s head of product for its API, Olivier Godement, emphasized the company’s attempts to streamline the fine-tuning procedure.
Details of prices
For firms on paid subscriptions, all developers have access to the fine-tuning capability. Users can select the base model from the fine-tuning dashboard and start customizing GPT-4o to get started.
Until September 23, 2024, OpenAI is giving away one million free training tokens per day to any organization. After that, inference is priced at $3.75 per million input tokens and $15 per million output tokens, while training costs $25 per million tokens.
GPT-4o micro fine-tuning, which offers 2 million free training tokens per day for the same duration, is an even more affordable choice.
Important aspects of fine-tuning GPT-4o
The process of fine-tuning enhances the model’s capacity to manage tone, structure, and intricate instructions, resulting in enhanced performance in certain applications.
The one- to two-hour process is usual, so customizations can be implemented quickly.
Strong safety measures, such as automatic assessments, have been put in place by OpenAI to stop the improper usage of optimized models.
What this implies
GPT-4o could be improved by a business to become a specialist chatbot for customer support that responds to in-depth product inquiries. This can be accomplished with as little as a few dozen samples of training data, which makes it useful and accessible for a variety of application scenarios.
The business claims that developers have been requesting this capability the most, and it represents a significant improvement in AI customization.
Data Confidentiality
According to OpenAI, users retain total control over their fine-tuned models and full ownership of their company data, including all inputs and outputs.
The business claimed to have included several levels of safety precautions to guarantee that optimized models are not abused, such as ongoing automatic safety assessments and monitoring to guarantee compliance with usage guidelines.
