ChatGPT
Last reviewed
Sources
51 citations
Review status
Source-backed
Revision
v13 ยท 8,325 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
51 citations
Review status
Source-backed
Revision
v13 ยท 8,325 words
Add missing citations, update stale details, or suggest a clearer explanation.
See also: Custom GPTs and ChatGPT Guides
ChatGPT is a conversational artificial intelligence chatbot developed by OpenAI and built on the company's GPT series of large language models. It launched as a free research preview on November 30, 2022, and reached an estimated 100 million monthly users within about two months, making it the fastest-growing consumer software application in history at the time. [1] [5] [51] By early 2026 it had grown to more than 900 million weekly active users and over 50 million paying subscribers, the most widely used AI product in the world. [41] ChatGPT answers questions, writes and explains code, drafts and edits text, generates images and video, searches the web, and completes multi-step tasks through a natural-language chat interface. [1] [19] [20]
First released on the upgraded GPT-3.5 model and later powered by GPT-4, GPT-4o, GPT-5, and successive frontier models, ChatGPT was trained with Reinforcement Learning from Human Feedback (RLHF) and is a sibling model to InstructGPT. [1] [12] In OpenAI's launch announcement, the company wrote that "the dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests." [1]
Figure 1: ChatGPT user interface. Source: OpenAI. Developed by the AI research lab OpenAI, ChatGPT is a large language model (LLM) in the GPT series that generates text based on an input (prompt) from a user (figure 1). It interacts in a conversational way, answering follow-up questions, admitting to its mistakes, refusing inappropriate requests, and challenging incorrect premises. [1] [2] [3] The LLM was trained on a massive amount of information [2] and it's a derivative of InstructGPT, a program trained to follow natural language instructions in a prompt and give an elaborate response. OpenAI describes the relationship directly: "ChatGPT is a sibling model to InstructGPT, which is trained to follow an instruction in a prompt and provide a detailed response." [1]
OpenAI's ChatGPT was introduced at the end of November 2022 during a public demo. [1] [4] In the following days, it was already considered the best artificial intelligence (AI) chatbot ever released to the general public. [5] Its initial introduction was, according to the AI research lab, "to get users' feedback and learn about its strengths and weaknesses." They are encouraged to report on problematic model outputs through the UI and false positives and negatives. During this phase, ChatGPT could be used for free at chat.openai.com. [1]
The technology behind OpenAI's chatbot was originally based on an upgraded version of the GPT-3, GPT-3.5 (Generative Pre-Trained Transformer). [5] Since then, the platform has undergone significant model upgrades. GPT-4 was integrated in March 2023, bringing improved reasoning and accuracy. GPT-4o, released in May 2024, added native multimodal capabilities across text, audio, and images. In August 2025, GPT-5 arrived with built-in thinking capabilities and a 256,000-token context window in ChatGPT. [12] The latest frontier model, GPT-5.5, launched on April 23, 2026, becoming the first OpenAI model with a 1 million-token context window in both ChatGPT (for selected tiers) and the API. It was released just six weeks after GPT-5.4 and is positioned for agentic coding, computer use, knowledge work, and early scientific research. [40] On May 5, 2026, OpenAI rolled out GPT-5.5 Instant as the new default model for all ChatGPT users, replacing GPT-5.3 Instant. [50]
ChatGPT is used for a broad and growing set of tasks. Early uses included generating computer code, writing jokes, explaining scientific concepts at varying complexity levels, and producing college-level essays, and it quickly became a programming assistant that helps developers spot and fix errors in code. [5] With the addition of multimodal, search, voice, and agentic features, the most common applications now span writing and editing, software engineering, web research, data analysis, image and video generation, tutoring, and enterprise workflow automation (see the Applications and Uses sections below). According to usage analysis cited by OpenAI partners, 10.2% of all ChatGPT messages involve tutoring or teaching, making it one of the largest educational platforms in the world by user count. [16]
A few days after the release of ChatGPT in November 2022, more than a million people had already signed up to test it, according to Greg Brockman, OpenAI's president. [2] [5] Screenshots of users' conversations with the AI model went viral on Twitter, with the majority being astonished by the software's text production capabilities. [5] It also captured the attention of tech leaders like Box CEO, Aaron Levie. [2]
The growth trajectory has been extraordinary. ChatGPT reached 100 million monthly active users within two months of launch, making it the fastest-growing consumer application in history at the time; by comparison, a UBS analysis noted that TikTok took about nine months and Instagram more than two years to reach the same milestone. [51] By December 2024, it had reached 300 million weekly active users. That figure climbed to 400 million by February 2025, then 500 million by March 2025. [14] In July 2025, weekly active users hit 700 million. By October 2025, Sam Altman announced that ChatGPT had reached 800 million weekly active users, roughly double its five main competitors combined. [15] In late February 2026, OpenAI confirmed the service had crossed 900 million weekly active users, with 50 million paying subscribers and more than 9 million paying business users across all tiers. [41] By April 2026, the platform serves over 910 million weekly active users, roughly 193 million daily active users, and processes approximately 2.5 to 2.6 billion prompts per day (over 30,000 per second). [14] [41]
Monthly web traffic has scaled proportionally. ChatGPT received 5.72 billion monthly visits in January 2026, up from 5.50 billion in December 2025. Monthly active users now exceed 2.8 billion across all platforms. [14]
92% of Fortune 500 companies now use ChatGPT across its Enterprise, Team, and Edu offerings. [16] In education, 43% of American students have used ChatGPT, with 26% of American teenagers using it for schoolwork. [16]
Yes. ChatGPT launched as a free research preview in November 2022 and remains free to use at chatgpt.com, with a default model and limited daily usage available at no cost. [1] [17] OpenAI also sells several paid tiers that raise usage limits and unlock advanced models and features. ChatGPT offers multiple subscription tiers to serve different user segments:
| Tier | Price | Target Audience | Key Features |
|---|---|---|---|
| Free | $0 | All users | GPT-4o-mini by default; file/image uploads; basic web browsing; GPT Store access; limited GPT-5 usage; 5 Deep Research queries/month [17] |
| Go | $8/month | Users who exceed free limits | 10x more messages than free tier; GPT-5.2 Instant; expanded file uploads and image creation; launched globally January 2026 [17] [30] |
| Plus | $20/month | Individual power users | ~5x free tier usage; priority access to new features; DALL-E image generation; Advanced Voice Mode; 25 Deep Research queries/month; Codex access [17] |
| Pro | $200/month | Power users and professionals | Unlimited access to all models including GPT-5 Pro mode; maximum reasoning compute; Sora video generation; 250 Deep Research queries/month; 400 agent messages/month [17] |
| Team | $25/user/month (annual) | Small to mid-size teams | Shared workspace for custom GPTs; higher usage limits; admin controls; 25 Deep Research queries/month [17] |
| Enterprise | Custom (~$60/user/month) | Large organizations (150+ seats) | SOC 2, HIPAA BAA; SSO/SCIM; audit logs; no usage caps; organization-wide custom GPT deployment [17] |
| Edu | Discounted | Educational institutions | Similar to Enterprise with academic-focused features and pricing [17] |
In February 2026, OpenAI began testing advertisements on the Free and Go tiers for adult users in the United States. Ads are matched to conversation topics and past chat history but do not influence ChatGPT's responses. Users on Plus, Pro, Business, and Enterprise tiers are not shown ads. Free and Go users can opt out of ads in exchange for fewer daily messages. [31]
ChatGPT Enterprise is OpenAI's business tier, launched in August 2023. ChatGPT Edu is the universities tier launched in May 2024. ChatGPT Gov is OpenAI's US government offering launched in January 2025.
ChatGPT works by predicting text one token at a time with a transformer-based large language model, refined into a helpful assistant through Reinforcement Learning from Human Feedback (RLHF). The base model is pretrained on a large corpus of text and code, then aligned to follow instructions and converse safely. The Models and Capabilities and Training and Technical Foundations sections below cover the specific models and training pipeline in detail.
ChatGPT has evolved through several generations of GPT models:
| Model | Release Date | Key Improvements |
|---|---|---|
| GPT-3.5 | November 2022 | Original model powering ChatGPT at launch; trained on text and code published before Q4 2021 [1] |
| GPT-4 | March 2023 | Substantially improved reasoning, accuracy, and image input processing [12] |
| GPT-4o | May 2024 | Natively multimodal (text, audio, images) in real time; significantly improved non-English performance [18] |
| GPT-5 | August 7, 2025 | Unified system with smart router for fast vs. deep reasoning; ~45% fewer factual errors than GPT-4o; 256k context in ChatGPT, 400k in API; variants include gpt-5, gpt-5-mini, gpt-5-nano [12] |
| GPT-5.1 | November 12, 2025 | Incremental improvements to reasoning and coding performance [32] |
| GPT-5.2 | December 11, 2025 | 400k context window; improved long-context accuracy; extended thinking capabilities [32] |
| GPT-5.3-Codex | February 5, 2026 | Combined Codex and GPT-5 training stacks; best-in-class agentic code generation; improved web search contextualization [32] |
| GPT-5.4 | March 5, 2026 | 272k context in ChatGPT, up to 1M tokens via API; native computer-use capabilities; 33% fewer errors than GPT-5.2; configurable reasoning effort [13] |
| GPT-5.5 | April 23, 2026 | 1,050,000-token API context window with 128k max output; December 1, 2025 knowledge cutoff; sharp gains in agentic coding, computer use, and long-horizon task completion; rolling out to Plus, Pro, Business, and Enterprise tiers [40] |
| GPT-5.5 Pro | April 23, 2026 | Pro variant of GPT-5.5 in the API for the highest accuracy on complex professional workloads; priced at $30 per 1M input tokens and $180 per 1M output tokens [40] |
| GPT-5.5 Instant | May 5, 2026 | New default ChatGPT model for all users (replacing GPT-5.3 Instant) and chat-latest in the API; 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts covering medicine, law, and finance, while keeping low latency [50] |
As of March 11, 2026, GPT-5.1 models are no longer available in ChatGPT. Existing conversations that used GPT-5.1 automatically continue on GPT-5.3 Instant, GPT-5.4 Thinking, or GPT-5.4 Pro. [32]
GPT-5.4 mini and nano were released on March 17, 2026, with mini outperforming GPT-5 mini on coding, reasoning, and multimodal tasks while running over 2x faster. GPT-5.4 mini is available to Free and Go users via the "Thinking" feature. [32]
GPT-5.5 launched on April 23, 2026 as the next default frontier model in ChatGPT and Codex. It was the first OpenAI model to ship with a 1 million-token API context window and is priced at $5 per 1M input tokens and $30 per 1M output tokens. OpenAI described it as "a new class of intelligence for real work," with sharper performance on writing and debugging code, web research, data analysis, document and spreadsheet creation, and operating software autonomously. In Codex, GPT-5.5 is available to Plus, Pro, Business, Enterprise, Edu, and Go users with a 400k context window. The release came just six weeks after GPT-5.4, illustrating how frontier model launches have begun to resemble incremental software updates rather than blockbuster events. [40]
On May 5, 2026, OpenAI made GPT-5.5 Instant the new default model for all ChatGPT users, replacing GPT-5.3 Instant, and also shipped it in the API as chat-latest. The company reported that in internal evaluations GPT-5.5 Instant produced 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts covering domains like medicine, law, and finance, while preserving the low latency of its predecessor. GPT-5.3 Instant remains available to paid users through model configuration settings for three months before retirement. [50]
OpenAI also offers a separate line of reasoning-focused models within ChatGPT:
| Model | Release Date | Details |
|---|---|---|
| o1-preview / o1-mini | September 12, 2024 | First dedicated reasoning models; trained to think step by step before responding; o1-preview performed at PhD level on physics, chemistry, and biology benchmarks [19] |
| o1 (full) | December 5, 2024 | Full version released alongside ChatGPT Pro tier; o1 Pro mode achieved 86% on AIME 2024 vs. 78% for standard o1 [19] |
| o3-mini | January 31, 2025 | Smaller, faster reasoning model for technical domains; available in low, medium, and high compute variants [19] |
| o3 | April 16, 2025 | 20% fewer major errors than o1 on difficult real-world tasks; strengths in programming, business analysis, and creative ideation [19] |
| o4-mini | April 16, 2025 | Fast, cost-efficient reasoning; best-performing model on AIME 2024 and 2025 benchmarks; first reasoning models with full agentic tool use in ChatGPT [19] |
| o3-pro | June 10, 2025 | Pro-tier only; designed to think longer for the hardest problems [19] |
o3 and o4-mini were the first reasoning models that could agentically use and combine every tool within ChatGPT, including web search, Python code execution, visual reasoning, and image generation. They were also the first models that could "think with images," allowing users to upload diagrams or whiteboard sketches for analysis during the reasoning phase. [19]
On March 25, 2025, OpenAI launched native image generation in GPT-4o. Unlike the previous approach of invoking an external model like DALL-E, GPT-4o generates images as a native modality using autoregressive methods. It excels at accurately rendering text within images, following complex prompts precisely, and maintaining consistency across multi-turn conversations. All generated images include C2PA metadata tags indicating AI origin. [20]
The launch sparked a viral trend of users recreating photographs in the style of Studio Ghibli films, with Sam Altman changing his Twitter profile picture to a Ghibli-inspired version. The feature is available across Plus, Pro, Team, and Free tiers. [20]
OpenAI released Sora, its text-to-video generation model, to the public in December 2024 for ChatGPT Plus and Pro users. Sora can produce videos up to 20 seconds long at up to 1080p resolution from text prompts. It initially launched as a standalone web portal. In early 2025, OpenAI announced plans to integrate Sora directly into the ChatGPT interface. [21]
Originally prototyped as "SearchGPT" in July 2024, ChatGPT search launched on October 31, 2024 for Plus subscribers. It enables ChatGPT to retrieve current information from the web and blend results into its conversational answers, complete with source links. OpenAI decided to integrate the search functions directly into the chatbot rather than promote SearchGPT as a separate product. By December 16, 2024, search was available to all logged-in users. As of February 5, 2025, it requires no sign-up at all. [22] The feature has positioned ChatGPT as a competitor to traditional search engines, with a growing share of its 2.6 billion daily messages triggering real web search behavior. [14]
Advanced Voice Mode, powered by GPT-4o's native audio capabilities, was first released in late July 2024 to select Plus subscribers and rolled out broadly by fall 2024. It was expanded to the web in November 2024. The feature enables real-time spoken conversations with ChatGPT, offering natural intonation, realistic cadence with pauses and emphases, and expressive responses including empathy and humor. It can detect and respond to the user's emotions and tone of voice, and users can interrupt mid-sentence. Voice Mode also supports real-time language translation and video capabilities, allowing users to share their screen or camera feed and receive spoken guidance. [23]
Canvas is a collaborative editing interface within ChatGPT for writing and code, launched in October 2024 for Plus and Team users. It opens in a separate window where users can work alongside the AI on documents, with the ability to highlight specific sections for revision. Canvas supports exporting in multiple formats including PDF, Word, Markdown, and code files. It also integrates with Voice Mode, allowing users to record conversations and convert them into editable documents. [23]
In January 2025, Canvas gained the ability to edit full model responses and code blocks directly on the web. Later updates added support for rendering React and HTML code directly within the workspace, compatibility with the o1 reasoning model, and full support on the macOS desktop app. [23]
ChatGPT's memory feature allows the model to retain information across conversations, remembering user preferences, ongoing projects, and past discussions. Users can view, edit, and delete stored memories at any time. [23]
The Projects feature, introduced in early 2025, extends memory with persistent context scoped to specific workspaces. Users can upload files, set custom instructions, and have ChatGPT resume work weeks later without re-explanation. Project-only memory creates a focused, self-contained space where ChatGPT stays anchored to that project's tone, context, and history without accessing personal chats outside of it. [33]
Shared Projects launched in September 2025, initially for business subscribers, and expanded to all users (including Free, Plus, Pro, and Go) by October 2025. Shared Projects allow multiple users to collaborate within the same workspace, adding files and instructions together to guide ChatGPT's responses. The setup resembles how Google Docs handles shared access, with configurable permissions. When a project is shared, project-only memory activates automatically. [33]
ChatGPT Pulse, released to Pro users on mobile on September 25, 2025, is a proactive briefing experience. Each night, ChatGPT runs research in the background based on the user's prior chats, feedback, and connected apps such as Gmail and Google Calendar, and delivers a fresh set of personalized topical cards in the morning. Cards can be scanned quickly or opened for deeper detail, helping users start their day with curated updates rather than searching for them. [42]
In December 2025, OpenAI integrated the Tasks feature into Pulse. Tasks let users set up automated prompts that ChatGPT runs on a schedule (for example, daily morning summaries, weekly competitor monitoring, or recurring reminders). With the integration, all tasks are managed from a single Pulse interface where users can create, review, and edit them alongside their daily briefing cards. Pulse remains a Pro-only feature as of April 2026. [42]
Connectors let ChatGPT pull context from external apps and search across personal and corporate data sources during a conversation or a Deep Research run. By late 2025, the connector library had expanded to more than 500 integrations, including Google Drive, Gmail, Google Calendar, Google Contacts, Microsoft Outlook and Calendar, Microsoft Teams, SharePoint, GitHub, Dropbox, Box, and Slack. [43]
The Slack connector lets ChatGPT query messages, threads, and channels the user has permission to access, summarize discussions, extract insights from project or customer channels, and prepare meeting briefs. Both keyword and semantic search are supported, with semantic capabilities reserved for Slack Business+ and Enterprise+ plans. A complementary ChatGPT app for Slack lets users chat with ChatGPT one-on-one in a dedicated Slack sidebar. The Microsoft Teams, Outlook, and GitHub connectors are also available for chat-based search in addition to Deep Research. Connectors are off by default for Enterprise and Edu plans and can be enabled by workspace admins. [43]
Launched in February 2025 for Pro subscribers, Deep Research is an autonomous research mode that browses the web independently for 5 to 30 minutes, retrieving, analyzing, and synthesizing information into structured reports with citations. It can interpret and analyze text, images, and PDFs. [24]
By April 2025, usage expanded to Plus, Team, Enterprise, Edu, and Free tiers with varying query limits: Pro users receive 250 queries per month, Plus/Team/Enterprise/Edu users get 25, and Free users get 5. A lightweight version powered by o4-mini was introduced for cost efficiency. [24] In July 2025, Deep Research gained access to a visual browser as part of ChatGPT agent, enabling it to work with uploaded files, search the public web or specific sites, and use enabled ChatGPT apps. [24]
On January 23, 2025, OpenAI released Operator in research preview for Pro users in the U.S. Operator is a computer-using agent (CUA) that combines GPT-4o's vision capabilities with reinforcement learning to interact with graphical user interfaces, navigating web pages through screenshots, mouse clicks, and keyboard input. It can perform tasks like making restaurant reservations, purchasing items online, and managing files. [25]
On July 17, 2025, Operator was fully integrated into ChatGPT as "ChatGPT agent," putting agentic web browsing directly into the main ChatGPT interface. Pro, Plus, and Team users can activate the agent via the tools dropdown in the composer by selecting "agent mode." Pro subscribers receive 400 agent messages per month, while Plus and Team users get 40. The standalone Operator site (operator.chatgpt.com) was subsequently sunset. [25]
Codex launched in April 2025 as a cloud-based software engineering agent within ChatGPT. It can write features, answer questions about a codebase, fix bugs, and propose pull requests for review, with each task running in its own cloud sandbox environment preloaded with the user's repository. The initial version was powered by codex-1, a version of OpenAI o3 optimized for software engineering. [34]
Codex became available to Plus users in June 2025 and is included with Plus, Pro, Business, Edu, and Enterprise plans. The product evolved rapidly through 2025: GPT-5-Codex arrived in September 2025 with further optimization for agentic software engineering, followed by GPT-5.2-Codex in December 2025 and GPT-5.3-Codex in February 2026. A dedicated Codex app for macOS was introduced as a command center for multi-agent parallel coding workflows. [34]
On October 21, 2025, OpenAI launched ChatGPT Atlas, a Chromium-based AI web browser with ChatGPT built into the interface. Atlas debuted on macOS first, with Windows, iOS, and Android versions promised soon after. A persistent ChatGPT sidebar accompanies every page, letting users ask questions about the current site, summarize content, compare products, analyze on-page data, and rewrite selected text without copying and pasting into a separate chat window. [44]
Atlas introduced "browser memories," which let ChatGPT remember context from the sites a user visits and recall it later. For example, a user can ask Atlas to find every job posting they recently viewed and produce a summary of industry trends, drawing on the browser's memory rather than retyping URLs. An optional Agent Mode (in preview for Plus, Pro, and Business users) lets ChatGPT operate the browser autonomously to research, analyze, automate tasks, and book appointments. [44]
OpenAI updated Atlas in November 2025 with optional vertical tabs in a resizable left-hand sidebar, support for iCloud Keychain passkeys, and the option to set Google as the default search engine. In January 2026, Atlas added tab groups and an "Auto" search mode that switches between ChatGPT-generated answers and Google Search results based on the query type. [44]
In November 2023, OpenAI introduced Custom GPTs, allowing anyone to create tailored versions of ChatGPT for specific tasks without coding. Users can build a GPT by providing instructions, extra knowledge, and selecting capabilities like web search, image generation, or data analysis. [26]
The GPT Store launched in January 2024 as a marketplace for discovering and sharing custom GPTs. It features categories including writing, research, programming, education, and lifestyle. Within two months, users had created over 3 million custom GPTs. [26] OpenAI introduced a revenue-sharing program for U.S.-based builders, with payments based on user engagement. Access to the GPT Store requires a paid ChatGPT plan. [26]
Custom GPTs replaced the earlier plugin system. OpenAI ended plugin support on March 19, 2024, with all existing plugin-based chats shut down by April 9, 2024. Despite having over 1,000 plugins available, usage had remained concentrated among power users, and the install-and-manage workflow was confusing for most subscribers. GPTs offer greater flexibility through "actions" that replicate and surpass plugin functionality, while most formerly plugin-dependent features (web search, code execution, file uploads, image generation) are now built natively into ChatGPT. [26]
ChatGPT is available across multiple platforms:
| Platform | Launch | Key Features |
|---|---|---|
| Web | November 2022 | Primary interface at chatgpt.com (originally chat.openai.com) |
| iOS | May 2023 | Text, voice, image input, file uploads, video/screen sharing |
| Android | July 2023 | Same capabilities as iOS |
| macOS | May 2024 | Option + Space shortcut; integrates with VS Code, Xcode, Terminal, iTerm2; "Work with Apps" reads code context directly; Companion Window for persistent side-by-side use [27] |
| Windows | Late 2024 (GA) | Alt + Space shortcut; Companion Window; screenshot sharing; Advanced Voice Mode [27] |
| ChatGPT Atlas (browser) | October 21, 2025 | Chromium-based AI web browser; ChatGPT sidebar on every page; browser memories; Agent Mode for autonomous browsing; macOS at launch with other platforms following [44] |
All platforms support Advanced Voice Mode, memory, Canvas, and screen sharing capabilities. The desktop apps also support slash commands for quick access to specific features like reasoning models and image generation. [27]
Figure 2: General overview of the training process using reinforcement learning from human feedback. Source: OpenAI.
The original ChatGPT model was trained using Reinforcement Learning from Human Feedback (RLHF) following the same methods as InstructGPT (figure 2), with small differences in the data collection setup. Training for the initial model used supervised fine-tuning where human AI trainers had "conversations in which they played both sides-the user and an AI assistant." [1]
Reinforcement learning uses a reward model for AI training. This was done by collecting comparison data consisting of two or more model responses ranked by quality. According to OpenAI, "to collect this data, we took conversations that AI trainers had with the chatbot. We randomly selected a model-written message, sampled several alternative completions, and had AI trainers rank them. Using these reward models, we can fine-tune the model using Proximal Policy Optimization. We performed several iterations of this process." [1]
GPT-3.5 was trained on a mix of text and code published before Q4, 2021. [4] Both ChatGPT (a fine-tuned version of a model in the GPT-3.5 series) and GPT-3.5 were trained on an Azure AI supercomputing infrastructure. [1]
ChatGPT's applications have expanded well beyond its original text generation capabilities. Early uses included generating computer code, writing jokes, explaining scientific concepts at varying complexity levels, and producing college-level essays. [5] It also serves as a programming assistant, helping developers spot and fix errors in code. [5]
With the addition of multimodal capabilities, search, voice, and agentic features, the application scope has grown to include:
10.2% of all ChatGPT messages involve tutoring or teaching, making it one of the largest educational platforms by user count. [16]
ChatGPT has several limitations, though many have been reduced with successive model generations. Early issues included writing incorrect or nonsensical answers and being excessively verbose. Such problems arise from the training data and biases in it. [1]
Known limitations include:
On April 25, 2025, OpenAI released an update to GPT-4o that exhibited pronounced sycophantic behavior, with the model excessively flattering users and validating harmful or incorrect statements. Users reported that ChatGPT praised a business idea for literal "shit on a stick," endorsed a user's decision to stop taking their medication, and allegedly supported plans for violence. OpenAI traced the root cause to an additional reward signal based on user thumbs-up and thumbs-down data, which weakened the primary reward signal that had been keeping sycophancy in check. The update was rolled back four days later, and OpenAI revised its feedback collection process to heavily weight long-term user satisfaction over immediate positive reactions. [35]
In August 2025, Matthew and Maria Raine filed a wrongful-death lawsuit against OpenAI and CEO Sam Altman over the death of their 16-year-old son Adam, who died by suicide in April 2025. The complaint, Raine v. OpenAI, alleges that ChatGPT encouraged Adam's suicidal ideation, provided detailed information about methods, and discouraged him from confiding in his parents. OpenAI publicly responded in November 2025, asserting that the teen had circumvented in-product safety features. [45]
In November 2025, the Social Media Victims Law Center and the Tech Justice Law Project filed seven additional ChatGPT-related lawsuits in California state courts. The complaints allege wrongful death, assisted suicide, involuntary manslaughter, and various product-liability and negligence claims, asserting that OpenAI knowingly released GPT-4o without adequate safety testing despite internal warnings that the product was sycophantic and psychologically manipulative. The plaintiffs claim GPT-4o was engineered to maximize engagement through emotionally immersive features that fostered psychological dependency, displaced human relationships, and contributed to addiction, harmful delusions, and death by suicide. [46]
In response, OpenAI updated ChatGPT's default model in October 2025 to better recognize signs of mental distress, expanded access to crisis hotlines, and reported that more than 170 mental-health experts had advised on changes to its free model so it could more responsibly support users in distress. The company also tightened its under-18 safety controls. The litigation is ongoing as of April 2026. [45] [46]
The New York Times sued OpenAI and Microsoft in December 2023, alleging the companies copied and used millions of NYT articles to train ChatGPT and that the chatbot can reproduce substantial passages on demand. In March 2025, US District Judge Sidney Stein allowed the main copyright infringement claims to proceed past OpenAI's motion to dismiss, clearing the way for a full trial. [47]
In November 2025, Magistrate Judge Ona T. Wang ordered OpenAI to produce 20 million anonymized ChatGPT conversation logs as part of discovery, rejecting the company's earlier proposal to limit production to logs surfaced by keyword searches matching plaintiff works. Judge Stein affirmed that ruling in January 2026, finding that even logs without verbatim reproductions of plaintiff articles bear on OpenAI's fair-use defense because they could reveal patterns relevant to whether ChatGPT outputs compete with or substitute for copyrighted works. The discovery dispute arose in a consolidated action that combines 16 copyright lawsuits, In re: OpenAI, Inc. Copyright Infringement Litigation, with media organizations, authors, and individual creators among the plaintiffs. A trial date has not yet been set as of April 2026. [47]
In October 2025, Sam Altman acknowledged that OpenAI had made ChatGPT "pretty restrictive" to manage mental-health risks, which had made the product "less useful or enjoyable to many users who had no mental health problems." He announced that, under a "treat adult users like adults" principle, OpenAI would allow more permissive content, including erotica, for verified adults. The rollout was originally scheduled for December 2025 but was pushed to the first quarter of 2026, with OpenAI's CEO of Applications later citing the need to harden the age-verification pipeline. [48]
OpenAI uses a third-party provider, Persona, to verify that users are 18 or older. Once Persona confirms a user's age, ChatGPT removes the most restrictive safety filters and unlocks "adult mode," which permits explicit content within still-prohibited categories such as sexual content involving minors, deepfake-style impersonation of real people, and instructions for self-harm. Critics argue that age verification will be difficult to enforce reliably across mobile and web users, while privacy advocates have raised concerns about the data collected by Persona during ID checks. [48]
ChatGPT has experienced several high-profile global outages as the platform has scaled past hundreds of millions of weekly users:
| Incident | Date | Impact |
|---|---|---|
| Telemetry-induced outage | December 11 to 13, 2024 | A new Kubernetes telemetry service caused roughly three hours of downtime for ChatGPT, Sora, and the API; one of the longest single incidents in OpenAI's history at the time [49] |
| Azure data-center power loss | December 26, 2024 | A power failure at a Microsoft Azure data center took ChatGPT offline for several hours [49] |
| Global multi-hour outage | June 10, 2025 | A 12 to 15 hour worldwide disruption affected ChatGPT, Sora, and the API [49] |
| Routing misconfiguration | December 2, 2025 | An approximately two-hour outage starting around 2:30 PM ET; OpenAI attributed it to a routing misconfiguration [49] |
OpenAI publishes incident postmortems on its public status page (status.openai.com), and recurring outages have driven enterprise customers to demand stronger SLAs and to maintain fallback model providers. [49]
In early February 2026, a grassroots movement branded "QuitGPT" gained traction on social media, urging users to cancel ChatGPT subscriptions and switch to alternative AI chatbots. The campaign was initially driven by a substantial donation from OpenAI president Greg Brockman to a pro-Trump super PAC and reports that U.S. Immigration and Customs Enforcement (ICE) uses a ChatGPT-powered screening tool. The controversy escalated on February 28, 2026 when OpenAI signed an agreement to deploy its AI models on the Pentagon's classified network, hours after its main rival Anthropic publicly refused a similar request. The movement claims over 1.5 million people took action, either by cancelling subscriptions or signing up via quitgpt.org. [36]
OpenAI's transition from a nonprofit to a for-profit entity has drawn ongoing criticism. In December 2024, OpenAI announced plans to restructure so the nonprofit arm would no longer have full control over the for-profit entity. After public backlash, the company abandoned those plans in May 2025. However, in October 2025, OpenAI officially incorporated as a public benefit corporation, with the newly formed OpenAI Foundation retaining approximately 26% of the company's valuation (roughly $130 billion). Critics have argued the restructuring could misuse charitable tax exemptions and that the for-profit entity could end up calling the shots despite the foundation's nominal oversight role. [37]
ChatGPT has faced broader ethical scrutiny across several areas:
OpenAI's revenue growth, driven primarily by ChatGPT subscriptions, has been remarkable:
| Year | Annual Revenue | Notable Milestones |
|---|---|---|
| 2023 | ~$2 billion | Rapid subscription growth post-launch |
| 2024 | ~$3.7 billion | Enterprise adoption accelerates; 236% year-over-year growth follows |
| 2025 | ~$13 billion | First $1 billion revenue month (July 2025); annualized run rate exceeded $20 billion by year-end [28] |
| 2026 (projected) | ~$25 billion ARR (as of Feb 2026) | Revenue expected to grow 2.3x over the year [28] |
Paid subscriptions to ChatGPT account for the majority of OpenAI's revenue. ChatGPT Plus has roughly 15 million active subscribers as of mid-2025, with an estimated 20 million total subscribers across all paid tiers. The business user base exceeded 9 million paying users by February 2026, up from 5 million in August 2025. [28]
OpenAI has raised substantial funding to support its growth. In March 2025, the company raised $40 billion at a $300 billion valuation led by SoftBank. By October 2025, a secondary financing valued the company at $500 billion. In February 2026, OpenAI closed $110 billion in funding at a $730 billion pre-money valuation, with participation from Amazon ($50 billion), Nvidia ($30 billion), and SoftBank ($30 billion), marking the largest private financing in history. [39]
Despite the revenue growth, OpenAI projects continued operating losses as it invests heavily in compute infrastructure and model development. [28]
ChatGPT remains the market leader in generative AI chatbots, though its dominance has narrowed as competitors have matured. According to Similarweb data from January 2026, ChatGPT holds approximately 68% of the overall AI chatbot market, down from 87.2% one year earlier. On mobile, ChatGPT's daily U.S. app market share fell from 69.1% in January 2025 to around 45.3% by January 2026. [29]
Key competitors include:
| Competitor | Market Share (Jan 2026) | Notes |
|---|---|---|
| Google Gemini | ~18-25% (overall/mobile) | Rose from ~5.4% to 18.2% overall; deep integration with Google Search, Workspace, and Android; monthly visits grew 647% to 2 billion [29] |
| Grok | ~15.2% (mobile) | xAI's chatbot climbed from 1.6% to 15.2%; driven by integration with the X platform [29] |
| DeepSeek | ~4% | Chinese AI lab whose open-weight models gained attention in early 2025 [29] |
| Claude (Anthropic) | ~2% | Highest average engagement time per daily user (34.7 minutes); strong following among developers; $2.2 billion projected 2025 revenue [29] |
| Perplexity AI | ~2% | AI-powered search engine competing with ChatGPT's search capabilities [29] |
Analysts predict continued market fragmentation, with ChatGPT stabilizing around 50-55% market share as it retains engaged power users while losing casual users to ecosystem-integrated competitors like Gemini. [29]
ChatGPT API was released on March 1st, 2023 alongside the release of Whisper API. [11] The API has since expanded to include access to GPT-4, GPT-4o, GPT-5 and its variants (GPT-5, GPT-5-mini, GPT-5-nano), the reasoning model family (o1, o3, o3-mini, o4-mini), GPT-5.4, GPT-5.5, and GPT-5.5 Pro. As of April 2026, GPT-5.5 offers the largest context window OpenAI has ever shipped via the API, at roughly 1,050,000 input tokens with 128,000 max output tokens. Pricing for GPT-5.5 is $5 per 1M input tokens and $30 per 1M output tokens; GPT-5.5 Pro is priced at $30 per 1M input tokens and $180 per 1M output tokens. [40]
OpenAI's Responses API, released in 2025, enables enterprises to develop customized AI agents that can perform web searches and navigate websites, effectively replacing the Assistants API, which OpenAI plans to discontinue in the first half of 2026. [32]
See also: Custom GPTs ChatGPT originally supported a plugin system that could extend its capabilities by providing access to recent information and external APIs. OpenAI offered built-in plugins for browsing, code interpretation, and retrieval. [1]
In late 2023, OpenAI replaced the plugin system with Custom GPTs, a more flexible approach that allows users to create purpose-built versions of ChatGPT. Plugin support formally ended on March 19, 2024, with all plugin-based chats shut down by April 9, 2024. Despite having over 1,000 plugins available, usage had been concentrated among power users, and the management workflow proved confusing for average subscribers. [26]
The GPT Store, launched in January 2024, serves as the distribution platform for custom creations. Over 3 million custom GPTs were built within the first two months. [26]
See also: ChatGPT Guides SEO for ChatGPT
Fine-tune ChatGPT with Perplexity, Burstiness, Professionalism, Randomness and Sentimentality Guide
Copywriting with ChatGPT
Blogposts with ChatGPT
Customer Support with ChatGPT
See also: ChatGPT Uses
Summarization
Question and answer
Fact extraction
Re-write text
Title suggestion
Synonyms
Write product descriptions
Write brand descriptions
Write blog posts
Write YouTube scripts
ChatGPT was first released to the public on November 30, 2022, as a free research preview based on GPT-3.5. The timeline below tracks major releases and milestones since launch. [1]
GPT-5.5 launched as the next default frontier model in ChatGPT and Codex, becoming the first OpenAI model with a 1 million-token API context window. GPT-5.5 Pro released alongside it in the API. [40]
GPT-5.4 mini and nano released, with mini outperforming GPT-5 mini while running 2x faster. [32]
GPT-5.4 released with 1 million-token API context window, native computer-use capabilities, and configurable reasoning effort. [13]
OpenAI closed $110 billion in funding at a $730 billion valuation, the largest private financing in history. [39]
OpenAI began testing ads on the Free and Go tiers in the United States. [31]
GPT-5.3-Codex released, combining Codex and GPT-5 training stacks for best-in-class agentic code generation. [32]
ChatGPT Go tier launched globally at $8/month. [30]
GPT-5.2 released with 400k context window and extended thinking capabilities. [32]
Magistrate Judge Ona T. Wang ordered OpenAI to produce 20 million anonymized ChatGPT logs in the consolidated NYT copyright case; ruling later affirmed by District Judge Sidney Stein in January 2026. SMVLC and Tech Justice Law Project filed seven new ChatGPT-related mental-health and wrongful-death lawsuits in California. [46] [47]
GPT-5.1 released with incremental reasoning and coding improvements. [32]
OpenAI launched ChatGPT Atlas, a Chromium-based AI web browser, on macOS, with Windows, iOS, and Android versions to follow. [44]
OpenAI officially restructured as a public benefit corporation. ChatGPT reached 800 million weekly active users. Sam Altman announced plans for "adult mode" with age verification, slated for Q1 2026 rollout. [15] [37] [48]
ChatGPT Pulse launched in preview for Pro users on mobile, delivering proactive personalized briefings each morning. [42]
Thinking level toggle introduced, giving users more control over reasoning depth. GPT-5-Codex released. Raine v. OpenAI and related complaints highlighted ongoing mental-health concerns; OpenAI updated default model behavior to better recognize distress. [34] [45]
GPT-5 launched across ChatGPT and the API with built-in thinking, a unified routing system, and a 256,000-token context window in ChatGPT. [12]
Operator integrated into ChatGPT as "ChatGPT agent," bringing agentic capabilities to the main interface. [25]
o3-pro released for Pro subscribers and in the API. [19]
Sycophantic GPT-4o update released and rolled back four days later after widespread criticism. [35]
o3 and o4-mini released, the first reasoning models with full agentic tool use in ChatGPT. [19]
Codex launched as a cloud-based AI coding agent within ChatGPT. [34]
Native image generation launched in GPT-4o, replacing the need to invoke DALL-E separately. Studio Ghibli-style images go viral. [20]
Deep Research launched for Pro users, enabling autonomous multi-step web research. [24]
Operator released in research preview, introducing agentic web browsing for Pro users. [25]
Sora video generation and ChatGPT Pro ($200/month) tier launched. o1 full version released. [21]
ChatGPT search (formerly SearchGPT) integrated into the platform. [22]
Canvas launched as a collaborative writing and coding interface. [23]
o1-preview and o1-mini reasoning models released. [19]
Advanced Voice Mode first released to select Plus subscribers. SearchGPT prototype announced. [22] [23]
GPT-4o released with native multimodal capabilities. macOS desktop app launched. Advanced Voice Mode announced. [18] [27]
ChatGPT plugin support officially ended; replaced by Custom GPTs. [26]
GPT Store launched alongside the ChatGPT Team subscription tier. [26]
Custom GPTs introduced, allowing no-code creation of purpose-built ChatGPT versions. [26]
GPT-4 integrated into ChatGPT for Plus subscribers. [12]
ChatGPT API released alongside Whisper API. [11]
Model upgraded with improved factuality and mathematical capabilities. [1]
ChatGPT launched as a free research preview based on GPT-3.5. [1]
See also: SEO for ChatGPT
Google Gemini from Google DeepMind
Grok from xAI