OpenAI Introduces Structured Outputs in API with GPT-4O Achieving 100% Reliability

TapTechNews August 7th news, OpenAI company yesterday (August 6th) released a press release announcing the introduction of Structured Outputs in the API.

OpenAI company has previously provided Structured Outputs, and users can specify response_format:{type:json_object} to request a valid JSON object, or use the function calling mechanism to request a response that conforms to a specific pattern.

OpenAI stated in the introduction that the Structured Output function can ensure that the response generated by the model always conforms to the JSON pattern provided by the user, so users don't have to worry about the model missing required keys or generating invalid enumeration values.

TapTechNews attached some advantages of Structured Output as follows:

Reliable type safety: No need to verify or retry responses with incorrect formats

Explicit rejections: Model rejections based on security can now be programmatically detected

Simplified prompts: Consistent formatting can be achieved without strongly worded prompts

The new model gpt-4o-2024-08-06 with Structured Outputs achieved a full 100% score in OpenAI's tracking and evaluation of complex JSON patterns; in contrast, gpt-4-0613 scored less than 40%.

Officials said that through Structured Output, gpt-4o-2024-08-06 achieved 100% reliability in its evaluation, perfectly matching the output pattern.

TapTechNews attached the reference address