OpenAI has officially introduced its latest advancement in artificial intelligence, the GPT-4o-2024-08-06 model. Announced yesterday, this new iteration brings significant enhancements to its API, most notably the inclusion of Structured Outputs support. This new feature ensures that outputs generated by the model align precisely with the JSON Schemas defined by developers, enhancing integration and usability.
In addition to this new capability, OpenAI has also reduced the pricing for API usage. The GPT-4o-2024-08-06 model is now available at a rate of $2.50 per million input tokens and $10.00 per million output tokens. This represents a 50% reduction in cost for input tokens and a 33% reduction for output tokens compared to the previous GPT-4o-2024-05-13 model.
Microsoft has announced that the GPT-4o-2024-08-06 model is now accessible through Microsoft Azure. However, Microsoft has yet to disclose the pricing details for this model on its platform, with updates expected to be provided soon.
The Structured Outputs feature is available in two variants:
User-defined JSON Schema: This option allows developers to specify the exact JSON Schema they wish the AI to follow. It is supported by both GPT-4o-2024-08-06 and the earlier GPT-4o-mini-2024-07-18 model.
More Accurate Tool Output (“Strict Mode”): This feature, available in a more limited format, enables developers to define specific function signatures for tool usage. It is supported by all models equipped for function calling, including GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, and GPT-4o models released from June 2023 onwards.
While OpenAI has not detailed performance enhancements specific to the GPT-4o-2024-08-06 model, early benchmarks suggest improvements over its predecessor. According to LiveBench, which evaluates language models with a focus on test set contamination and objective metrics, the GPT-4o-2024-08-06 model achieved an average score of 56.71, an increase from the 54.63 average of the GPT-4o-2024-05-13 model. Furthermore, the ZeroEval Leaderboard currently places the GPT-4o-2024-08-06 model at the top with an average score of 88.5275. However, it has not yet been listed in the LMSYS Chatbot Arena, where the Gemini-1.5-Pro-Exp-0801 model currently leads with a record score of 1299.
The launch of the GPT-4o-2024-08-06 model marks another milestone in OpenAI’s quest to advance AI capabilities, offering developers both greater control and cost-efficiency. As real-world applications begin to adopt this new model, its true impact and performance will be closely observed.
Related topics:
Why Hasn’t Sora Been Released Yet?Â