OpenAI is constantly updating its models and pricing, and today marks another significant milestone. The company has recently unveiled a range of new models, along with a price drop for API access, which is of particular interest to developers and also acts as a signal for future customer options.
GPT-3.5 Turbo, the most widely used model, known for its use in ChatGPT, has become the industry standard – if responses are not on par with those of ChatGPT, why bother? It is also a popular API, offering lower costs and faster performance compared to GPT-4 for various tasks. Thus, paying users will be delighted to hear that input prices have reduced by 50%, while output prices have decreased by 25%, now at $0.0005 per thousand input tokens and $0.0015 per thousand output tokens.
As more people explore using these APIs for text-heavy applications, such as analyzing extensive papers or books, the tokens begin to accumulate rapidly. With the emergence of open-source or self-managed models that match OpenAI’s performance, the company needs to retain its customers. Thus, they continuously lower prices, aided by advancements in streamlining and infrastructure.
GPT-3.5 Turbo now boasts a new model version, 0125 (reflecting today’s date), featuring “various improvements.” However, the company has not disclosed these updates in detail. The previous version, 0613, makes it slightly surprising that the company has not mentioned more.
Furthermore, GPT-4 Turbo has a new preview model for API usage, also named 0125, with an intriguing fix:
This model now more comprehensively completes tasks, such as code generation, compared to the previous preview model, aiming to reduce instances of “laziness” where the model fails to complete a task.
Perhaps the model has learned too much from the developers whose practices it has absorbed and emulated. Who knows, maybe AI is already on the brink of quitting. That would be good for AI indeed!
Do keep in mind that this is currently in preview mode. However, GPT-4 Turbo with vision (GPT-4 V) will officially launch in the next few months.
In addition to these exciting updates, OpenAI has also released a few new and enhanced text embedding models, geared towards its technical users. Moreover, the company has introduced a new version of its free moderation API that identifies potentially harmful text – version 007 is now available for those experimenting with using it for their moderation needs.