Superintelligence News — Developers now have the opportunity to fine-tune GPT-4o, enabling higher performance and accuracy tailored to specific use cases. As of today, fine-tuning for GPT-4o is available to all developers, marking a significant advancement in model customization. This new feature responds to one of the most requested capabilities by developers, allowing them to adjust the model’s structure and tone or follow complex, domain-specific instructions.
Unlocking Customization with Fine-Tuning
Fine-tuning offers the flexibility to adapt GPT-4o to a wide range of applications, from coding to creative writing. With this launch, developers can fine-tune the model using custom datasets, achieving high performance with just a few dozen examples. Through September 23, every organization can utilize 1M training tokens per day for free, making it easier than ever to enhance their models.
To begin fine-tuning, developers can visit the fine-tuning dashboard, select the gpt-4o-2024-08-06 base model, and start customizing. The training cost is set at $25 per million tokens, with inference costing $3.75 per million input tokens and $15 per million output tokens. A mini version, GPT-4o mini, is also available, offering 2M free training tokens per day through September 23.
Success Stories: Achieving State-of-the-Art Performance
Several companies have already leveraged GPT-4o’s fine-tuning capabilities to achieve state-of-the-art results. For instance, Cosine’s Genie, an AI software engineering assistant, has shown remarkable performance improvements on the SWE-bench benchmark. Fine-tuning allowed Genie to autonomously identify and resolve bugs, build features, and refactor code with higher accuracy, leading to a state-of-the-art (SOTA) score of 43.8% on the SWE-bench Verified benchmark.
Similarly, Distyl, an AI solutions provider for Fortune 500 companies, ranked 1st on the BIRD-SQL benchmark, the leading text-to-SQL benchmark. Distyl’s fine-tuned GPT-4o achieved an impressive 71.83% execution accuracy, excelling in tasks such as query reformulation, intent classification, and SQL generation.
Prioritizing Data Privacy and Safety
Fine-tuning also ensures that developers maintain full ownership of their business data, including all inputs and outputs, safeguarding against misuse. The models are equipped with layered safety mitigations and continuous automated evaluations to ensure adherence to usage policies.
As GPT-4o’s fine-tuning capabilities continue to evolve, the possibilities for customization and enhanced application performance are vast. Developers interested in exploring more options can reach out to the team for further support and guidance.
Fine-tuning GPT-4o provides a powerful tool for developers looking to enhance performance and accuracy in their applications, making it a valuable asset for businesses and organizations across various domains.
