OpenAI only unveiled its latest model, GPT-4o, a few weeks ago, but the AI giant has already rolled out a few updates.
The company unveiled the multimodal GPT-4o in a demo in May. And while it uses the GPT-4 model, it is optimized to be faster and more efficient.
As part of the update to GPT-4o, users will see two significant changes, including the ability to structure outputs - which should make life easier for developers - and a price drop reportedly aimed to compete with Google Gemini.
What you need to know about structured outputs in GPT-4o
The biggest change is the introduction of "Structured Outputs" in the API, which allows developers using GPT-4o to direct the format of the end data — in particular by matching JSON Schemas provided by users, a keyword system for defining data, structures and more to ensure interoperability and consistency.
"Generating structured data from unstructured inputs is one of the core use cases for AI in today’s applications," OpenAI said in a blog post.
Currently, developers have been working around the inability for large language models (LLM) to output necessary formats, the company said.
"Structured Outputs solves this problem by constraining OpenAI models to match developer-supplied schemas and by training our models to better understand complicated schemas," the blog post added.
OpenAI noted that the Structured Outputs system had been developed to fit within the company's existing safety policies, meaning it will still refuse unsafe requests.
In a blog post, Microsoft's Azure team said the new output feature focuses on "enhancing productivity" for developers.
OpenAI said the feature will work on all models that support tools, including GPT-4o, the August update, and all models including 'gpt-4-0613' and 'gpt-3.5-turbo-0613' and later.
A juicy price drop
OpenAI's price drop appears to be its latest move in a price war with Google, this time focused on larger models. To be clear, this isn't the pricing for individuals or businesses to access ChatGPT and the like, but to make use of the models themselves via an API.
OpenAI said the updated version of GPT-4o is to be priced at $2.50 per 1 million input tokens and $10.00 per 1 million output tokens. That's cheaper than the previous iteration by nearly half for inputs and a third for outputs, and now undercuts Google's Gemini 1.5 Pro model.
Earlier this year, Google unveiled its Gemini 1.5 Flash model, which was priced lower than rival systems at $0.35 per million input tokens and $1.05 per million output tokens, according to reports.
When OpenAI unveiled the GPT-4o Mini model, it was 60% cheaper than the standard OpenAI model and cheaper than Google Gemini 1.5 Flash too, at $0.15 cents per million input tokens and 60 cents per million output tokens.
Google responded last week by slashing the price of Gemini 1.5 Flash to $0.075 per million input tokens and $0.3 per million output tokens, according to Neowin.
