27-1-2024 (SAN FRANCISCO) OpenAI, known for its innovative AI models, has unveiled several adjustments to its models and pricing, marking another milestone in the AI industry’s evolution. While primarily of interest to developers, these changes also hint at future consumer options.
At the forefront of OpenAI’s offerings is GPT-3.5 Turbo, widely recognized as an industry standard, notably through its flagship application, ChatGPT. This model stands out for its popularity and efficiency, often preferred over GPT-4 due to its lower cost and faster performance on various tasks. To the delight of paying users, OpenAI has reduced input prices by 50% and output prices by 25%, now priced at $0.0005 per thousand tokens in and $0.0015 per thousand tokens out.
As developers explore the potential of these APIs for text-intensive applications such as analyzing papers or books, the significance of these price reductions becomes evident. Moreover, as competing open-source or self-managed models strive to match OpenAI’s performance, the company must ensure customer retention, hence the gradual price adjustments. These adjustments also reflect ongoing efforts to streamline models and enhance infrastructure.
The latest iteration of GPT-3.5 Turbo, version 0125, introduces “various improvements,” although specifics remain undisclosed, a departure from the detailed updates of previous versions like 0613.
In addition, GPT-4 Turbo debuts a new preview model for API use, also version 0125, addressing issues such as code generation comprehensiveness, aiming to minimize instances of incomplete tasks, or “laziness,” observed in previous models. While still in preview mode, GPT-4 Turbo with vision (GPT-4 V) is slated for general availability “in the coming months.”
Alongside these updates, OpenAI introduces new and enhanced text embedding models, catering primarily to technical users. Moreover, the company releases a new iteration of its free moderation API, designed to identify potentially harmful text. Those interested in utilizing this feature for moderation purposes can explore version 007.
