The original AI model behind ChatGPT will live on in your favorite apps

OpenAI has released its GPT‑3.5 Turbo API to developers as of Monday, bringing back to life the base model that powered the ChatGPT chatbot that took the world by storm in 2022. It will now be available for use in several well-known apps and services. The AI brand has indicated that the model comes with several optimizations and will be cheaper for developers to build upon, making the model a more efficient option for features on popular applications, including Snapchat and Instacart. 

Apps supporting GPT‑3.5 Turbo API

OpenAI detailed some of the companies that will use the GPT‑3.5 API to update their popular applications. Many of the apps will use the API to incorporate AI chat features into their interfaces. Snap Inc. recently announced a My AI feature for its Snapchat+ subscription tier. It will serve as a custom chatbot for users, offering recommendations and text editing, among other unique functions for Snapchat’s 750 million monthly users. Quizlet is a learning platform used by over 60 million students globally, which will integrate the API as an AI tutor that can interact with students based on their topic and level of study. Instacart collaborates with over 75,000 retail locations to provide shopping experiences. Its “Ask Instacart” API integration, set to launch later this year, will allow users to ask functional questions, such as how to make a certain recipe, to assist with shopping needs. Shopify is an e-commerce app that allows its 100 million shoppers to easily find their desired products and brands. The addition of the GPT‑3.5 API will enable a shopping assistant that will provide personalized, prompt-based recommendations of the app’s extensive product database. 

GPT‑3.5 history and updates

ChatGPT 3.5 was OpenAI’s first publicly available AI model, with the introduction of ChatGPT in November 2022. It powered the chatbot’s free tier with various updates from its inception until July 2024, when it was succeeded by GPT-4o mini. The current GPT-4o mini model remains in use at the base mode for all ChatGPT tiers. During its time, ChatGPT 3.5 was especially known for its hallucinations, or making up responses when it didn’t have the answers readily available. This was primarily because the model was trained on data up to September 2021 and could not answer questions beyond that point. 

However, OpenAI indicated that the GPT‑3.5 Turbo API has been formulated for developers to use to create beyond chat-based products. The company added that all updates made available will be a stable version, which it recommends developers opt for use. The current GPT‑3.5 Turbo API is also cheaper for developers to use at $0.002 per 1k tokens and is 10x cheaper than prior GPT‑3.5 models.

Whisper API shows open-source upgrades 

OpenAI also discussed its upgrade of the Whisper API, a speech-to-text model that was first developed as an open-source model and made available to developers in September 2022. The brand noted that the model was originally hard to run but has now been optimized for faster performance. Whisper API works by transcribing the source language into English. It works with several audio formats, including, m4a, mp3, mp4, mpeg, mpga, wav, and webm. 

An app that currently supports the Whisper API is Speak, an AI-powered language learning app available in South Korea, which is the premier app for learning English in the country. The Whisper API update is intended to expand support for Speak globally. The app promises to assist users with achieving fluency in language learning, with open-ended practice and accurate feedback.

Notably, OpenAI is providing an update to the Whisper AI API– a model that has always been open source in its library, after the sudden popularity of the Chinese AI brand DeepSeek. The company’s CEO, Sam Altman, stated in a Reddit AMA in early February that OpenAI was “on the wrong side of history” and that the company needed to reconfigure its open-source strategy.

Additionally, OpenAI’s chief product officer, Kevin Weil, also stated during the AMA session that there was potential for the company to make its older, less cutting-edge models open-source. He didn’t give any specifics about which model could be used for an open-source project at that time; however, Whisper API appears to be a good example of what he had in mind.  

GPT‑3.5 Turbo isn’t the only ghost model 

OpenAI has been making many changes to model availability recently. Over several months, the brand has had a focus on releasing several reasoning models, which are better at logical thinking and can show the thought process behind results. As said, the ChatGPT base model is currently a reasoning model, and OpenAI allows full access to several other reasoning models through its paid subscription tiers. With several reasoning model options available, OpenAI recently announced its decision to retire the GPT-4 large language model (LLM), which powered the ChatGPT Plus tier before GPT-4o mini. The model will be available in the ChatGPT interface until April 30.  

OpenAI detailed that the GPT-4 LLM will still be available as an API for developers. While it remains unknown how OpenAI will use the GPT-4 API moving forward, seeing what the company did with GPT‑3.5 Turbo might give an idea of what’s in store. 

Notably, when OpenAI first announced GPT-4 in March 2023, the company detailed its several business integrations of the LLM, including the language learning app Duolingo, the education website Khan Academy, the Danish mobile app Be My Eyes, the payment app Stripe, the AI agent company Auto-GPT, and the AI assistant You.com, in addition to its major collaboration with Microsoft for its Copliot AI assistant.