By Emmanuel Ogbodo
OpenAI has launched a new feature that enables businesses and developers to fine-tune its most advanced model, GPT-4o, using their own data for $25, starting September 24, 2024. This new capability allows companies to customise the model’s responses to better align with specific requirements, such as adjusting structure, tone, and following complex, domain-specific instructions.
Olivier Godement, OpenAI’s Head of Product for its API, emphasised the company’s focus on making the fine-tuning process accessible and straightforward. “We’ve been extremely focused on lowering the bar, the friction, the amount of work it takes to get started,” said Godement.
The fine-tuning feature is available to all developers and companies on paid plans. Users can start by visiting the fine-tuning dashboard, selecting the base model, and customising GPT-4o according to their needs. OpenAI is offering 1 million free training tokens per day for each organisation until September 23, 2024. Post that, training will cost $25 per million tokens, with inference priced at $3.75 per million input tokens and $15 per million output tokens.
For those seeking a more cost-effective option, OpenAI has also introduced GPT-4o mini fine-tuning, which offers 2 million free training tokens per day during the same period. The fine-tuning process typically takes just one to two hours, allowing businesses to quickly implement and deploy customised solutions.
This feature significantly enhances the model’s ability to manage structure, tone, and complex instructions, leading to improved performance in specialised applications. For instance, a company could fine-tune GPT-4o to develop a specialised customer service chatbot capable of answering detailed product questions with minimal training data, sometimes requiring only a few dozen examples.
OpenAI has implemented robust safety protocols, including continuous automated evaluations, to prevent the misuse of fine-tuned models. Importantly, companies retain full control over their fine-tuned models, with complete ownership of their business data, including all inputs and outputs.
This fine-tuning capability has been one of the most requested features by developers and represents a significant step forward in AI customization, making it more accessible and effective for a wide range of applications.
