It’s not uncommon for companies to choose poor names for their products, particularly in the technology sector. OpenAI is a prime example of this. The AI startup’s sheer volume of models and products has led to a confusing and perplexing naming and numbering convention.
Initially, OpenAI’s AI models had relatively straightforward names, such as GPT-1, GPT-2, and GPT-3. However, things began to get complicated over time. The introduction of unexpected variants, such as GPT-4o (which the company chose not to call GPT-5, a title reserved for a much more advanced model expected in the future), added to the confusion.
Here’s the current list of OpenAI models:
- GPT-1 (June 2018): This is OpenAI’s first version of the pre-trained generative AI model.
- GPT-2 (February 2019): With GPT-2, OpenAI’s AI model expanded significantly, boasting 1.5 billion parameters. Initially, access was very limited to prevent misuse.
- GPT-3 (June 2020): This marked the introduction of the first full-fledged large language model with 175 billion parameters, though access remained restricted.
- GPT-3.5 (March 2022): OpenAI made advancements in generative AI, but users didn’t see this model in action until a few months later. OpenAI finally released the GPT-3.5 Turbo, which was cheaper and optimized for production use.
- GPT-4 (March 2023): This model significantly improved its performance and input data capacity. Shortly after, the company released the more affordable and faster GPT-4 Turbo version.
- GPT-4o (May 2024): This model introduced notable improvements over GPT-4, including exceptional quality in synthesized speech and the addition of new vision capabilities.
- GPT-4o mini (July 2024): OpenAI started releasing lighter and more efficient versions of its models, such as GPT-4o-mini. The “mini” models are the new “Turbo” and offer more modest performance. Additionally, they’re accessible even in the free version of ChatGPT.
- o1 (September 2024): This is OpenAI’s first reasoning model. The o1 model takes time to generate responses and review its answers, marking a significant advancement from previous models.
- o1-mini (September 2024): This version is more affordable and efficient than the o1 model, although it has limited capacity. Notably, there was no “o2” model because a British telecommunications services provider already used the name.
- o3 (December 2024): When the second version of OpenAI’s reasoning model was introduced, it substantially improved performance. At the same time, the company released a more limited and resource-efficient version.
- o3-mini (January 2025): This modest and efficient version provided widespread access to the advanced reasoning capabilities of the o3 model.
AI Models Everywhere
These are OpenAI’s main AI models, but there are actually more. For instance, OpenAI also offers DALL-E 3 for image generation, TTS for speech synthesis, and Whisper for audio transcription.
Additionally, some models have different versions, such as “preview” editions (preliminary versions) or “latest” versions (the most recent ones, like chatgpt-4o-latest, currently used in ChatGPT).

What’s more, each model often has multiple versions defined by the dates it was “compiled.” These versions can reflect improvements. For example, according to OpenAI, the GPT-4o model currently has five different versions.

Aside from their release dates, these versions generally have minimal differences. The newer versions tend to perform slightly better and may also support longer response lengths in terms of maximum token output. The pricing for these models varies based on their capabilities. You can find more information in the official OpenAI documentation.
This multitude of versions and variants has understandably created confusion for users and even for OpenAI itself, which is attempting to manage the availability of both free and paid subscriptions across these models.
OpenAI CEO Sam Altman has even acknowledged this confusing situation himself. After the presentation of GPT-4o-mini, he said in an X post that a redesign of the company’s model naming scheme is certainly necessary.
However, no changes have been made as of now. The situation has become even more complicated with certain reasoning models, whose names have been unclear from the outset. For instance, OpenAI has models named o1 and o3 but not o2.
ChatGPT Has Several Versions, Too
The most well-known OpenAI service is ChatGPT, a generative AI chatbot that revolutionized the tech world when it was launched in November 2022.
ChatGPT serves as an interface for accessing and utilizing the company’s AI models. Initially based on GPT-3.5, there are now multiple versions of ChatGPT available, depending on the subscription you choose:
- ChatGPT (free): The free version is currently based on GPT-4o mini and offers a limited number of queries. After exceeding that limit, users must wait several hours before they can use it again. Additionally, ChatGPT offers two special modes. On one hand, the Search mode accesses the Internet to find and summarize information for users. On the other, the newly announced Reason mode uses the o3-mini model for a limited number of queries.
- ChatGPT Plus ($20/month): Launched in February 2023, this paid version provides broader access to the chatbot. Users can interact more extensively than with the free model, upload more documents, and generate additional images using DALL-E 3. It also offers more o3-mini queries and access to the GPT-4o voice mode.
- ChatGPT Pro ($200/month): This ambitious model grants priority (though often limited) access to the most advanced OpenAI models. It includes extensive access to GPT-4o voice mode and limited access to the o3 reasoning model, along with advanced features such as Operator, Deep Research, and Sora for video generation.
- ChatGPT Team ($30/month): Launched in January 2024, this subscription is designed for small companies that wish to share access to the model while ensuring data privacy. It offers higher limits than ChatGPT Plus and provides priority access to some of the new OpenAI service models and options.
- ChatGPT Enterprise: Introduced in August 2023, this plan is tailored for large corporations and offers special access to GPT-4o. It emphasizes data privacy for the companies using the service. Pricing is flexible and negotiated directly with OpenAI, and it offers priority access to new models and service options.
Image | Ishmael Daro
Related | Faced With the Rise of DeepSeek, Sam Altman Admits His Vision About Open Source Was Wrong
View 0 comments