ChatGPT is a conversational chatbot built by OpenAI on its GPT family of large language models, first released as a free research preview on November 30, 2022. GPT stands for generative pre-trained transformer: a model first trained to predict the next token in a text, then tuned with human feedback to follow instructions. As of October 24, 2026, OpenAI sells it in a free tier and paid plans from USD 8 to USD 500 per month.
Key takeaways
- ChatGPT was fine-tuned from a model in the GPT-3.5 series using reinforcement learning from human feedback, according to OpenAI's November 30, 2022 announcement.
- OpenAI itself listed plausible-sounding but incorrect answers as a known limitation at launch.
- The company began in 2015 as a nonprofit. Since October 28, 2025, a nonprofit OpenAI Foundation controls a for-profit public benefit corporation, OpenAI Group PBC.
- Plans and prices change often, so every figure below carries the date it was read.
What GPT means and how the chatbot is trained
A GPT model is a transformer, a neural network design, trained on very large amounts of text to predict the next token in a document. The 2020 paper "Language Models are Few-Shot Learners," submitted to arXiv on May 28, 2020 by 31 authors, described GPT-3 as a model with 175 billion parameters that could perform tasks from instructions and examples given in plain text, without retraining. That paper also noted that the model struggled on some datasets and raised concerns about training on large web collections.
ChatGPT added a second stage. OpenAI's announcement says human trainers wrote conversations playing both the user and the assistant, and that trainers then ranked several model replies to build a reward model. The company used that ranking to fine-tune the model with a method called Proximal Policy Optimization. The starting point was a model from the GPT-3.5 series that finished training in early 2022.
A dated timeline
The table lists events that the company or the cited papers date themselves.
| Date | Event | Source type |
|---|---|---|
| December 11, 2015 | OpenAI announced as a nonprofit AI research company | OpenAI post |
| May 28, 2020 | GPT-3 paper submitted to arXiv | Research paper |
| November 30, 2022 | ChatGPT released as a free research preview | OpenAI post |
| March 14, 2023 | GPT-4 announced, accepting image and text inputs and producing text | OpenAI post |
| October 28, 2025 | Recapitalization completed, creating OpenAI Group PBC under the OpenAI Foundation | OpenAI post |
| January 16, 2026 | ChatGPT Go made available worldwide | OpenAI post |
The GPT-4 technical report, posted to arXiv on March 15, 2023, gave no parameter count, training data details, or architecture specifics. Outside observers therefore cannot check how the model was built from the report alone.
Limits the developer documents itself
The 2022 announcement is unusually direct about weaknesses. It says ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers, and gives three reasons: reinforcement training has no source of truth, making the model more cautious makes it decline questions it could answer, and supervised training can mislead because the ideal answer depends on what the model knows. It also says the model is sensitive to rephrasing, is often verbose, usually guesses what an ambiguous question meant rather than asking, and will sometimes respond to harmful instructions or show biased behavior despite filters.
Those statements date from 2022, and the models have changed since. OpenAI's own page now notes that the post "introduced ChatGPT in 2022" and that the product has evolved substantially. The limitations of that era are still a reasonable starting list of questions to ask about any chatbot answer, though the frequency of each problem in current models is a separate, measurable matter that this article does not quantify.
Plans and prices read on October 24, 2026
OpenAI's Help Center describes a Plus plan at USD 20 per month. Its January 16, 2026 post gave the United States price of Go as USD 8 per month, noted that Go pricing is localized in some markets, and listed Pro at USD 200 at that time. The Pro tiers page, updated within the past few days, now lists Pro 100 at USD 100, Pro 200 at USD 200, and Pro 500 at USD 500 per month, and says new Pro 200 subscriptions carry a lower usage allowance than before, with existing subscribers keeping the older allowance through October 29, 2026. API use is billed separately from consumer plans. Prices outside the United States, taxes, and Canadian-dollar amounts were not verified.
Corporate structure in brief
OpenAI describes itself as founded in 2015 as a nonprofit with the mission of ensuring artificial general intelligence benefits all of humanity. It says its for-profit business was established in 2019, and that on October 28, 2025 the nonprofit, renamed the OpenAI Foundation, kept control of the for-profit, now a public benefit corporation called OpenAI Group PBC. The company says the Foundation holds equity valued at approximately USD 130 billion, and that the change followed discussions with the attorneys general of California and Delaware. These are the company's statements, and independent valuations were not checked.
Checking a chatbot answer before relying on it
- Ask for the source of any figure, then open the source and confirm the number.
- Rephrase the question once and compare the two answers, since the announcement notes that phrasing changes results.
- Treat legal, medical, and financial answers as a starting point for a conversation with a professional.
- Avoid pasting personal or confidential data unless the plan's data terms have been read.
- Note the model name and date shown on the screen, because both change.
FAQ
Is ChatGPT the same thing as GPT?
No. GPT refers to the family of underlying language models. ChatGPT is the chat product built on models from that family, with additional training to follow instructions in a dialogue.
Who owns the company behind ChatGPT?
According to OpenAI, the OpenAI Foundation, a nonprofit, controls OpenAI Group PBC, the for-profit public benefit corporation formed in the October 28, 2025 restructuring. Stakes held by other parties were not verified here.
Is ChatGPT free?
A free tier exists. The Go, Plus, and Pro plans add higher limits and more features, and the terms of each plan are set by OpenAI and can change.
Can the answers be trusted?
Not without checking. OpenAI's own launch notes list incorrect but plausible answers as a known limitation, and later models have not removed the need to verify important claims.
Related reading
- GPT-3 and the language model behind early chatbots
- What GPT-3 chatbots can and cannot do
- Building a GPT chatbot for Telegram
- Choosing AI chatbot solutions
More in AI Tools and Industry.
Last verified: October 2026
Sources
- Introducing ChatGPT, OpenAI, November 30, 2022
- Built to benefit everyone: OpenAI's updated structure, OpenAI, October 28, 2025
- Introducing OpenAI, OpenAI, December 11, 2015
- Language Models are Few-Shot Learners, Brown et al., arXiv, 2020
- GPT-4 Technical Report, OpenAI, arXiv, 2023
- GPT-4, OpenAI, March 14, 2023
- Introducing ChatGPT Go, now available worldwide, OpenAI, January 16, 2026
- What is ChatGPT Plus? and About ChatGPT Pro tiers, OpenAI Help Center, read October 24, 2026