What is ChatGPT

What it is

ChatGPT is a conversational AI chatbot developed by OpenAI. It generates text, and in later versions speech and images, in response to user prompts, using large language models from OpenAI's GPT family. It is the product that turned large language models from a research curiosity into a mass-market tool, and the reason the current AI era has a clear starting date.

OpenAI launched ChatGPT on November 30, 2022 as a free research preview, built on GPT-3.5 and fine-tuned with reinforcement learning from human feedback, the same approach used for InstructGPT. Human trainers played both sides of sample conversations to teach the model to follow instructions in a dialogue format, and a reward model trained on ranked responses steered it toward answers people preferred. The launch needed no marketing budget. It reached a million users in five days and a hundred million monthly active users within two months, the fastest-growing consumer application in history at the time. By early 2026, Wikipedia reported 900 million weekly active users.

How it works

At its core, ChatGPT wraps a large language model in a conversational interface with memory of the current session. The model predicts the next tokens of a response given the conversation history and any instructions, which produces fluent dialogue, explanations, code, and structured output. The chat format was the breakthrough as much as the model: it gave ordinary people a way to use a technology that had previously required an API key and code.

Around that core, OpenAI layered tools the model can invoke. Web browsing grounds answers in current information. A code interpreter runs Python and analyzes uploaded files. Image generation, voice conversations, file uploads, and deep research features that compile multi-source reports all became part of the product. Custom GPTs let users build tailored versions with their own instructions and knowledge, and share them with others. The underlying model has been upgraded repeatedly, from GPT-3.5 to GPT-4, GPT-4o, the o-series reasoning models, and GPT-5 generation models, with the app routing different tasks to different models behind the scenes.

Key details

OpenAI runs ChatGPT on a freemium model. The free tier covers everyday use with limits. ChatGPT Plus costs $20 per month, per OpenAI's help center, and adds higher limits, priority access during busy periods, faster responses, and expanded features including voice conversations, image generation, file analysis, and deep research tools. API usage is billed separately from subscriptions, a distinction that confuses new users.

Higher tiers exist for heavier use. Pro plans at $100 and $200 per month offer much larger usage allowances aimed at researchers and professionals. A Go plan at $8 per month launched in some markets. Business and Enterprise plans add administrative controls, single sign-on, compliance features, and contractual protections around data use, including no training on business data. Plan details shift with rollouts, so OpenAI's own pricing page is the reliable source for what each tier includes today.

Limitations

The well-known weakness is hallucination: the model can generate incorrect or nonsensical answers that sound plausible, because it predicts likely text rather than verifying facts. That makes it unreliable for anything where accuracy matters unless a human checks the output. Its training data carries biases that surface in responses, and the model can reflect those biases back with the same confidence as its facts.

The tooling around it has raised its own issues, from academic dishonesty to misinformation to malicious code generation. The ethics of its development remain contested, particularly the use of copyrighted material in training data, which has drawn lawsuits from authors and publishers.

And the product moves fast enough that specifics age quickly. Model availability, limits, and plan details change with rollouts; features appear in some regions before others; the model lineup behind the interface gets swapped without announcement. For a product used by hundreds of millions of people, the only durable advice is to check what your account can actually do today rather than trusting a description written months ago.

The everything app

ChatGPT kept absorbing features until the chat interface became a platform. Custom GPTs let anyone package instructions, knowledge, and tools into a shareable assistant, and the GPT Store gave them distribution. Memory gave the model continuity across sessions, remembering preferences and context the way a human assistant would. Projects organized chats, files, and instructions around longer work. Canvas provided a workspace for writing and code alongside the conversation.

The tool integrations went deeper than most users realize. Beyond browsing and code execution, the model gained the ability to generate images, analyze spreadsheets, read PDFs, and conduct multi-step research tasks autonomously. Voice mode turned the phone app into something closer to a conversational companion, with the low-latency audio that GPT-4o enabled. Each addition made the product harder to describe and harder to leave; the switching cost of a tool that holds your files, your history, and your custom assistants is substantial.

That expansion is also the source of the product's strangest property: nobody uses the same ChatGPT. The free user, the Plus subscriber, the team workspace, and the API developer interact with different model lineups, different limits, and different features, all under one name. The product is less a single chatbot than a family of services sharing a brand. For a wiki article, that means every specific claim about ChatGPT needs a date and a tier attached, because the only constant in its history has been change.