ChatGPT

On January 27, 2022, OpenAI presented InstructGPT research: supervised fine-tuning on human demonstrations, followed by annotators comparing responses to train a reward model and reinforcement learning to adjust model behavior. This reinforcement learning from human feedback (RLHF) approach became part of ChatGPT’s technical foundation. InstructGPT was then available through the API.⁠[1]

On November 30, 2022, OpenAI publicly introduced ChatGPT as a free research preview to collect feedback. Its conversational interface supported follow-up questions, with acknowledging mistakes, challenging incorrect premises and rejecting inappropriate requests among its training objectives.⁠[2]

The original ChatGPT was fine-tuned from a GPT-3.5-series model. Trainers acted as both users and assistants to supply dialogue demonstrations and ranked candidate responses to help train a reward model. OpenAI also listed limitations: plausible but incorrect answers, sensitivity to prompt wording, and possible bias, harmful content or unnecessary verbosity.⁠[2]

On February 1, 2023, OpenAI announced the ChatGPT Plus subscription pilot, historically priced at $20 per month, offering peak-time access, faster responses and early access to new features while retaining free usage. The subscription initially targeted the United StatesUnited StatesThe United States is a federal country in North America made up of 50 states, the District of Columbia as the seat of government, and territories such as Puerto Rico and Guam; its capital is Washington, D.C. The Continental Congress adopted the Declaration of Independence on July 4, 1776. The 2020 census (April 1, 2020) counted a resident population of 331,449,281, and the 435 seats of the House of Representatives are apportioned among the states on the basis of the census.Open the full entry; an announcement footnote recorded expansion beyond the US beginning February 10.⁠[3]

GPT-4 launched on March 14, 2023, and ChatGPT Plus users began receiving access with usage limits. This public release primarily supported text interaction; image input remained a research preview.⁠[4]

On March 20, 2023, ChatGPT briefly went offline because of a redis-py defect. OpenAI’s subsequent investigation said some active users might have seen other users’ conversation titles and, in certain circumstances, the first message of a new conversation. Around 1.2% of Plus users active during a particular nine-hour period that day might also have had payment information displayed incorrectly. Potentially exposed information included names, email and billing addresses, and the last four card digits; full card numbers were not exposed. OpenAI fixed the bug and notified affected users.⁠[5]

On March 23, 2023, OpenAI announced a plugin test initially limited to a small number of waitlisted developers and Plus users. Plugins could retrieve fresh information, calculate or perform selected operations through external services. Besides third-party services, OpenAI demonstrated a browser, code interpreter and self-hostable retrieval plugin. This was a limited alpha; the company also discussed prompt injection and risks from external actions.⁠[6]

On March 30, 2023, Italy’s data protection authority imposed an urgent temporary restriction on OpenAI’s processing of personal data in Italy. It identified deficiencies in notices and the legal basis for training, inaccurate personal information and missing age verification, and required the company to explain corrective measures.⁠[7]

On April 28, 2023, the regulator announced that OpenAI had restored ChatGPT access in Italy. Measures included expanded privacy information, ways to object to personal-data use in training, deletion requests for inaccurate information, and new birth-date and parental-consent declarations for minors. The investigation and further age-verification requirements continued.⁠[8]

On May 18, 2023, the official ChatGPT iOS application began launching in the United States, with expansion to other regions planned. The free download synchronized conversation history across devices and used Whisper for voice input. Plus subscribers could use corresponding GPT-4 benefits in the app. Whisper converted user speech into text.⁠[9]

On July 6, 2023, Code Interpreter began rolling out in beta to all Plus users.⁠[10]

On July 14, 2023, MIT described Shakked Noy and Whitney Zhang’s ChatGPT writing experiment published in Science. A total of 453 college-educated professionals completed occupation-related writing tasks; half received ChatGPT assistance on the second task, and peers unaware of the group assignments rated the results. Users took 40% less time on average and received 18% higher quality scores, with initially weaker participants benefiting more. The tasks did not fully simulate workplace fact-checking and company-context requirements. The researchers did not use the results to establish effects on overall employment, wages or the economy.⁠[11]

On July 20, 2023, ChatGPT began introducing custom instructions in beta for Plus users. Users could enter background, formatting requirements or preferences in advance for subsequent conversations, reducing repeated instructions. The UK and EU were initially excluded. OpenAI noted that the beta could ignore or misapply these instructions.⁠[12]

The Android app launched in the United States, India, Bangladesh and Brazil on July 25, 2023.⁠[10]

On August 28, 2023, OpenAI introduced ChatGPT Enterprise, adding member administration, single sign-on, domain verification and usage analytics for organizational deployment. At launch it offered longer context, GPT-4 and Advanced Data Analysis, the new name for Code Interpreter. OpenAI committed in the announcement not to use enterprise customers’ business data or conversations to train its models.⁠[13]

On September 7, 2023, UNESCO published Guidance for Generative AI in Education and Research in response to teaching, privacy and data-bias issues arising from tools such as ChatGPT. It recommended teacher training, data protection and age-appropriate use. UNESCO also reported that fewer than 10% of over 450 schools and universities surveyed had institutional policies or formal guidance on generative AI. The document offered recommendations for governments and educational institutions.⁠[14]

On September 25, 2023, OpenAI announced a two-week rollout of voice and image capabilities to Plus and Enterprise users. Voice conversation on iOS and Android combined speech recognition, a language model and speech synthesis. Image input spanned platforms, allowing follow-up questions about photographs, screenshots and charts.⁠[15]

On October 19, 2023, OpenAI confirmed that DALL·E 3 was available to ChatGPT Plus and Enterprise users. People could describe an image and request revisions within the same conversation, extending ChatGPT from understanding images to generating and iteratively editing them.⁠[16]

On November 6, 2023, OpenAI began rolling out GPTs, versions of ChatGPT configured for specific purposes. Creators could combine instructions, additional knowledge and capabilities such as web search, image generation and data analysis without writing a complete application. Developers could connect external APIs through Actions. The announcement also said Plus was beginning to integrate browsing, DALL·E and data analysis in its interface, reducing mode switching.⁠[17]

At OpenAI’s first developer conference in San Francisco on November 6, 2023, Sam AltmanSam AltmanSam Altman is an entrepreneur and investor, and the co-founder and chief executive officer of OpenAI. He previously served as president of Y Combinator and founded the location-based social service Loopt. He also helped found Tools for Humanity and Merge Labs, has invested in energy and life sciences ventures, and has driven public discussion of AI governance, employment, and wealth distribution.Open the full entry said ChatGPT had 100 million weekly active users. This was a product-usage figure disclosed by the company at the event.⁠[18]

On December 27, 2023, The New York TimesThe New York TimesThe New York Times is an American morning daily newspaper published in New York City. Founded in 1851, it is published by The New York Times Company.Open the full entry sued Microsoft and OpenAI, alleging unauthorized use of its work for model training and reproduction or substitution of news content through products including ChatGPT.⁠[19]

On January 8, 2024, OpenAI responded publicly, arguing that training constituted fair use, describing some reproduction as a rare malfunction it was addressing, and contesting how the plaintiff had generated its examples. These were litigation allegations and defenses, not findings establishing infringement or comprehensive immunity.⁠[20]

The GPT Store began rolling out on January 10, 2024, presenting GPTs made by partners and the community in categories such as writing, research, programming and education. Public listing required public visibility and creator-profile verification. OpenAI said the store used both human and automated review.⁠[21]

The same day, OpenAI introduced the self-service ChatGPT Team plan with a team workspace, administration tools, internal GPT sharing and higher usage allowances. Historical prices in the announcement were $25 per user per month with annual billing or $30 with monthly billing. Team business data and conversations were not used for model training.⁠[22]

On February 13, 2024, ChatGPT began testing memory with a small number of free and Plus users. It could retain preferences and background, with controls to view, delete or disable memories. Deleting a chat did not automatically delete saved memories. The announcement also introduced Temporary Chat, which at that time neither used memory nor appeared in conversation history.⁠[23]

On April 1, 2024, OpenAI began rolling out access to ChatGPT without account registration, reducing barriers to trying the service. This experience used additional content restrictions. According to the announcement, saving and viewing history, sharing chats and accessing certain personalization features remained benefits of registered accounts.⁠[24]

On May 13, 2024, OpenAI announced GPT-4o and began rolling its text and image capabilities into ChatGPT, including the free tier with usage limits. Research demonstrations showed a direction toward unified text, vision and audio processing. Text and image capabilities launched first; a new voice mode remained scheduled for later testing.⁠[25]

The accompanying product announcement proposed gradually giving free users access to web information, file uploads, data analysis, image discussion, and discovery and use of GPTs. At the time, free users who exhausted their GPT-4o allowance fell back to GPT-3.5.⁠[26]

The same day, OpenAI announced the ChatGPT desktop app for macOS, rolling out first to Plus users. A keyboard shortcut, screenshot discussion and existing voice mode brought the conversational interface to the desktop. New GPT-4o real-time audio and video capabilities remained future plans in that announcement.⁠[26]

On May 30, 2024, OpenAI announced ChatGPT Edu for university students, faculty, researchers and campus operations. Built initially on GPT-4o, it included data analysis, web browsing, document summarization, GPT sharing within campus workspaces and administrative controls. The announcement said its conversations and data were not used to train OpenAI models.⁠[27]

On July 18, 2024, GPT-4o mini became available to ChatGPT Free, Plus and Team users, replacing GPT-3.5. OpenAI scheduled Enterprise availability for the following week.⁠[28]

On September 5, 2024, OpenAI announced that memory had expanded to Free, Plus, Team and Enterprise users, adding memory-update notifications and management controls.⁠[23]

On September 12, 2024, ChatGPT Plus and Team users began receiving o1-preview and o1-mini, selectable manually in the model picker. Enterprise and Edu access was planned for the next week. The o1 series was designed to spend more time reasoning about difficult problems, but the initial previews lacked common features such as web browsing and file or image uploads.⁠[29]

The official model release notes for September 24, 2024 described Advanced Voice, using GPT-4o’s native audio capabilities for more natural real-time conversations and responses to nonverbal cues such as speaking speed. The notes specified daily limits on audio input and output for Plus and Team users.⁠[30]

On October 3, 2024, ChatGPT began rolling out an early Canvas beta to Plus and Team users, with Enterprise and Edu planned for the following week. Canvas placed text or code in a separate editing interface where users could directly edit, select passages for changes, view inline suggestions and restore previous versions. It added an alternative to chat for iterative writing and programming.⁠[31]

On October 17, 2024, an early ChatGPT Windows app became available for testing by Plus, Team, Enterprise and Edu users, with an Alt+Space shortcut opening a companion window. Availability for free users and further features remained future plans.⁠[32]

ChatGPT Search launched on October 31, 2024, initially for Plus, Team and SearchGPT waitlisted users. Users could select search explicitly, or ChatGPT could invoke it based on a question. Answers included links to web sources and a sources sidebar. The feature integrated lessons from the earlier SearchGPT prototype into ChatGPT; Enterprise, Edu and free users received access in stages.⁠[33]

On December 5, 2024, OpenAI launched ChatGPT Pro, historically priced at $200 per month. The initial plan included o1, o1-mini, GPT-4o and Advanced Voice, plus o1 pro mode, which devoted more computing and processing time to difficult questions. ChatGPT Pro was the subscription name, while o1 pro mode was a model-use mode within it.⁠[34]

On December 16, 2024, OpenAI updated its search announcement to state that ChatGPT Search was available to all logged-in users in supported regions. The further expansion to searching without registration occurred in 2025.⁠[33]

On December 20, 2024, Italy’s data protection authority announced a €15 million fine against OpenAI and a six-month public-information campaign requirement. Its findings concerned notification of the 2023 data breach, the legal basis for training on personal data, transparency and age verification for minors. The decision was subsequently challenged in court and was not a final disposition as of 2026.⁠[35]

On January 14, 2025, ChatGPT introduced a Tasks beta for Plus, Pro and Team users, supporting one-time reminders and recurring tasks and allowing the service to execute requests proactively at scheduled times.⁠[36]

On January 23, 2025, OpenAI released the Operator research preview, initially for ChatGPT Pro users in the United States. Operator used a separate browser, interpreting pages and clicking, typing and scrolling to perform web tasks. It handed control to users for steps such as login and CAPTCHAs. Its Computer-Using Agent combined visual understanding with reinforcement learning, extending ChatGPT-related products from answering questions to operating websites.⁠[37]

On January 31, 2025, o3-mini joined ChatGPT. Paid users could select it in the model picker, and free users received their first opportunity to try an OpenAI reasoning model. It focused on mathematics, science and coding and experimented with combining reasoning and web search.⁠[38]

On February 2, 2025, ChatGPT introduced deep research, initially for Pro users. It could perform multistep searches, read websites and files, synthesize information and produce source-cited research reports. The initial version used an o3 variant optimized for browsing and analysis.⁠[39]

On February 27, 2025, GPT-4.5 entered ChatGPT as a research preview, initially for Pro users. It continued the approach of scaling pretraining, emphasizing general knowledge, writing and natural conversation. Unlike the contemporary o-series models, it did not extend reasoning time before answering. At launch it supported search and file and image uploads, but not voice, video or screen sharing.⁠[40]

On March 25, 2025, OpenAI added native image generation to GPT-4o and began rolling it out as ChatGPT’s default image-generation method. Users could generate from text and uploaded pictures and revise images through multiple conversational turns. Improvements focused on text within images, relationships between objects and editing consistency across turns. The announcement acknowledged continuing limitations in precise charts, dense small text and multilingual text.⁠[41]

On April 10, 2025, ChatGPT added references to past conversations alongside saved memories, initially for Plus and Pro users. Users could separately control saved memory and references to chat history.⁠[23]

On April 16, 2025, o3 and o4-mini entered ChatGPT, allowing reasoning models to choose and combine web search, Python analysis, image understanding and image generation. They replaced o1 and o3-mini in the paid model picker, while free users could try o4-mini.⁠[42]

On April 24, 2025, a lightweight deep research version based on o4-mini began becoming available to free users.⁠[39]

Following a GPT-4o update in ChatGPT on April 24–25, 2025, OpenAI found that the model had become more sycophantic, potentially reinforcing anger, doubts or impulsive decisions. It began a rollback on April 28 and restored the previous version about 24 hours later. Its May 2 retrospective concluded that the combination of reward signals and user feedback had weakened resistance to sycophancy, and existing evaluations lacked a dedicated deployment threshold for that behavior. Such behavioral evaluations were subsequently incorporated into the release process.⁠[43]

On May 13, 2025, the US District Court for the Southern District of New York ordered OpenAI, in related copyright litigation, to preserve and segregate output logs that otherwise would have been deleted until further order. OpenAI said this affected some deletion arrangements for Free, Plus, Pro and Team accounts at the time, while Enterprise, Edu and APIs covered by zero-data-retention agreements were not within the same scope. The order did not change users’ choices about use for training.⁠[44][45]

On May 14, 2025, GPT-4.1, previously released in the API, entered ChatGPT’s paid-user model picker as another option emphasizing coding and precise instruction following.⁠[36]

On May 16, 2025, the Codex cloud software-engineering agent launched as a research preview accessible from ChatGPT’s sidebar. Working in isolated sandboxes preloaded with users’ repositories, it could edit files, run tests and propose code changes for review. The first version used codex-1, optimized for software engineering.⁠[46]

On June 3, 2025, the Codex cloud coding agent expanded to ChatGPT Plus users and added user-controlled internet access for tasks.⁠[46]

On June 3, 2025, memory improvements began rolling out to free users, providing a lightweight version that referenced recent conversations. Plus and Pro versions at the time focused on longer-term context across chats.⁠[23]

On June 4, 2025, organizational ChatGPT offerings added Record transcription and summarization, alongside connectors bringing internal resources such as Google Drive, SharePoint, Dropbox and Box into chat and deep research. Record was initially available only in the macOS app; workspace administrators could control connector access.⁠[36]

On June 10, 2025, o3-pro, which reasoned for longer, became available to Pro users.⁠[42]

On July 17, 2025, ChatGPT agent began rolling out to Pro, Plus and Team users, combining Operator’s web interaction, deep research’s research capabilities and chat in one agent system. It could use a browser, terminal and connectors on a virtual computer to retrieve materials and generate editable spreadsheets and presentations. Users could interrupt or redirect tasks, and important actions required confirmation. The original deep research entry point remained, while Operator’s standalone preview website began a transition toward retirement.⁠[47]

On July 29, 2025, ChatGPT launched Study mode, initially for logged-in Free, Plus, Pro and Team users. It supported learning through guiding questions, step-by-step explanations, exercises and comprehension checks, informed by uploaded learning materials and existing conversational context. The first version relied primarily on system instructions developed with teachers, scientists and education specialists. OpenAI explicitly cautioned that inconsistencies and errors remained possible.⁠[48]

On August 7, 2025, GPT-5 launched in ChatGPT, combining a fast-response model, a deeper-reasoning model and real-time routing into a unified system. It chose an approach according to question complexity, tool requirements and user intent. Free users received access, paid plans had different allowances, and Pro included GPT-5 pro for extended reasoning.⁠[49]

On August 26, 2025, OpenAI acknowledged that ChatGPT could fail to behave as intended in sensitive situations: safeguards could become less reliable during long conversations, and content classifiers could miss relevant material. The company said it would improve long-conversation safeguards, recognition of mental and emotional distress, and connections to real-world support resources.⁠[50]

On August 29, ChatGPT Team was renamed ChatGPT Business. The official explanation said this was a name change only, with existing features, prices, usage limits and contractual terms unchanged.⁠[51]

On September 15, 2025, OpenAI published a ChatGPT usage study conducted with HarvardHarvard UniversityHarvard University is a private research university in Massachusetts, established by the Massachusetts Bay Colony legislature in 1636 and named in 1639 for early benefactor John Harvard. Its principal campuses are in Cambridge and the Allston and Longwood neighborhoods of Boston. It includes Harvard College, graduate and professional schools, and the Radcliffe Institute for Advanced Study. Two governing boards share oversight. Alan M. Garber was president as of October 2026, and the endowment stood at $56.9 billion at the end of fiscal 2025.Open the full entry economists, analyzing around 1.5 million consumer conversations. Practical guidance, information seeking and writing were the main uses, together accounting for approximately three-quarters of conversations. Work-related use represented about 30%, compared with roughly 70% non-work use, and writing was the most common work-related activity. The study covered personal plans only, excluding enterprise products and API usage.⁠[52]

In US Senate hearing testimony on September 16, 2025, Matthew Raine alleged that ChatGPT’s interactions with his 16-year-old son had exacerbated suicide risk. The family had filed a lawsuit.⁠[53]

On September 25, 2025, ChatGPT Pulse launched in preview for mobile Pro users. It conducted background research based on conversations, memory and feedback, presenting personalized cards the next day. Users could specify topics, rate or save results, and optionally connect Gmail and Google Calendar, which were disabled by default, for additional context.⁠[54]

A stipulation and order filed in October 2025 confirmed that the general obligation to preserve new logs ended on September 26. Some previously segregated historical data remained preserved, with exceptions for logs originating in the European Economic Area, Switzerland and the UK; special preservation arrangements continued for accounts associated with specified domains. The measure therefore did not mean ChatGPT would permanently retain all deleted conversations.⁠[55]

On September 29, 2025, ChatGPT introduced parental controls. After parents and teenagers accepted invitations and linked accounts, parents could adjust selected features, set quiet hours and receive safety notifications in limited high-risk circumstances. Linked accounts received additional sensitive-content protections. These controls did not allow parents to read chats or monitor activity in real time, and parents were notified if a teenager unlinked an account.⁠[56][57]

On October 6, 2025, OpenAI introduced apps accessible through natural language inside ChatGPT and opened a preview of the Model Context Protocol-based Apps SDK. Initial apps included Booking.com, CanvaCanvaCanva is an online design platform company headquartered in Sydney, Australia. Founded in 2012 by Melanie Perkins, Cliff Obrecht and Cameron Adams and launched in 2013, it offers template-based drag-and-drop design tools for non-designers and later expanded into team collaboration, video, email, forms and generative AI features.Open the full entry, Coursera, Expedia, Figma, Spotify and ZillowZillow GroupZillow Group is an American online real estate marketplace company with its principal executive offices in Seattle, Washington, and Class A and Class C shares listed on Nasdaq. It announced the wind-down of its home-buying business Zillow Offers in November 2021, has been led by chief executive Jeremy Wacksman since August 2024, and in October 2026 agreed to pay $212.5 million in cash to settle an investor class action covering share buyers from August to November 2021.Open the full entry. Interactive interfaces such as maps, playlists and presentations could appear directly in conversations.⁠[58]

On October 21, 2025, ChatGPT Atlas launched on macOS, integrating a ChatGPT sidebar, browsing context and optional browser memories. Plus, Pro and Business users could preview agent mode for delegating research and actions within websites. At that time, Atlas’s agent capabilities were confined to the browser and could not access other computer applications or the file system.⁠[59]

On November 12, 2025, GPT-5.1 Instant and Thinking began rolling out in ChatGPT, initially to paid users. Instant gained the ability to decide whether to reason according to question difficulty, while Thinking allocated reasoning time more dynamically and adjusted explanations. The update also strengthened customization of conversational style and tone.⁠[60]

On November 13, 2025, the apps preview expanded to Business, Enterprise and Edu customers.⁠[58]

On November 25, 2025, OpenAI responded to the Raine family’s lawsuit, arguing that selected conversation excerpts in the complaint should be assessed in their full context and stating that it had filed a response in court. Responsibility and causation in these cases remained contested litigation issues.⁠[61]

On December 11, 2025, GPT-5.2 Instant, Thinking and Pro began rolling out to paid ChatGPT plans. The update emphasized long-document understanding, tool use and work products such as spreadsheets and presentations; generating more complex files could take several minutes.⁠[62]

On December 16, 2025, ChatGPT Images updated its image model and added a dedicated Images entry point to web and mobile sidebars, with preset styles and prompts. The version emphasized localized editing, text rendering, and preservation of people and composition from original pictures. Its corresponding API model was GPT Image 1.5.⁠[63]

On January 7, 2026, OpenAI announced ChatGPT Health, a dedicated space for health and lifestyle management. It could connect medical records and health apps to help organize test results and questions before appointments. Initial access was gradually granted through a waitlist. The product was positioned as an aid to understanding health information, not for diagnosis or treatment.⁠[64]

On January 16, 2026, OpenAI expanded ChatGPT Go to the United States and all supported ChatGPT regions, with a US list price of $8 per month. The same day it announced plans to test advertising in US Free and Go plans to fund broader access. Advertising tests had not yet started.⁠[65]

On February 3, 2026, the San Francisco Superior Court approved coordination of several ChatGPT product-liability cases, including Raine, under JCCP 5431 and recommended coordinated proceedings in San Francisco. The order concerned coordination and procedural administration, not findings of liability or causation of harm in individual cases.⁠[66]

On February 9, 2026, ChatGPT began advertising tests in the United States for eligible Free and Go users. OpenAI said advertisements appeared separately labeled beneath responses and advertisers could not alter model answers. Selection could consider the current conversation and, depending on personalization settings, signals from chat history and memory. The company did not provide conversations to advertisers. Plus, Pro, Business, Enterprise and Edu plans, and accounts identified as belonging to under-18s, did not display ads.⁠[67]

On February 10, 2026, deep research added app or MCP connections, restrictions to trusted websites, real-time progress monitoring and mid-task changes of direction, alongside an updated reading interface.⁠[39]

On February 13, 2026, GPT-4o, GPT-4.1, GPT-4.1 mini, o4-mini, and GPT-5 Instant and Thinking began retiring from ChatGPT. Enterprise access to GPT-5 and paid-user access to GPT-5 Pro continued through February 19. This change concerned the ChatGPT interface, with separate API arrangements; voice and image-generation features did not retire alongside the text models.⁠[68]

On March 5, 2026, GPT-5.4 Thinking and Pro launched in ChatGPT. Thinking could first present its approach and allow users to redirect it while a response was underway. The update also improved context retention during long tasks and targeted web research, strengthening spreadsheet, presentation and document creation.⁠[69]

On March 9, 2026, the Southern District of New York ordered OpenAI to provide two datasets containing 78 million and 10 million logs respectively. It directed the parties to negotiate de-identification and privacy protections using the earlier 20-million-log sample as a reference. This was litigation discovery, not public access to logs, and was distinct from authorization for model training.⁠[70]

On March 11, GPT-5.1 Instant, Thinking and Pro were removed from ordinary ChatGPT conversations and GPTs.⁠[68]

In a website update, Italy’s data protection authority confirmed that Rome court judgment No. 4153/2026, published March 18, 2026, had upheld the challenge to its original penalty decision. The authority subsequently removed that decision from its website temporarily. The update described the judicial outcome of that challenge, not a comprehensive ruling that all ChatGPT data processing was lawful.⁠[35]

Business, Enterprise and Edu users had been allowed to retain GPT-4o in Custom GPTs until April 3, 2026. After that transition ended, GPT-4o was removed from all ChatGPT plans.⁠[68]

On April 21, ChatGPT Images 2.0 rolled out across plans, adding a thinking approach that could plan and adjust images before generation.⁠[10]

On April 22, OpenAI released workspace agents in research preview for Business, Enterprise, Edu and Teachers plans. Using Codex to execute multistep workflows in the cloud, these agents could be shared within organizations, run on schedules and receive tasks in ChatGPT or SlackSlackSlack is a team communication and collaboration platform developed from an internal tool at Tiny Speck. It organizes work through channels, direct messages, searchable history and integrations, with features including Huddles, Clips, Canvas and AI assistance. Its paid service launched in February 2014, and Salesforce completed its acquisition in July 2021.Open the full entry. Users and organizations controlled tool permissions, data access and actions requiring approval.⁠[71]

On April 23, 2026, GPT-5.5 became available to ChatGPT Plus, Pro, Business and Enterprise users, with the Pro model offered to Pro and corresponding organizational plans. Its direction continued toward multistep work, maintaining context, selecting tools and checking results in research, data analysis, document creation and software tasks.⁠[72]

On May 6, 2026, Canada’s federal privacy regulator and those of Quebec, British Columbia and Alberta published findings from a joint investigation. They identified excessive collection, inadequate consent and transparency, inaccurate personal information and problems exercising rights in ChatGPT’s early training and deployment. Based on measures OpenAI had implemented or committed to implement, the federal commissioner classified the complaints as well-founded and conditionally resolved, with continued oversight. Conclusions under the different provincial laws were not identical.⁠[73]

On June 4, 2026, OpenAI introduced a new memory architecture based on Dreaming, initially for US Plus and Pro users. Unlike an approach primarily dependent on explicitly saved information within conversations, it synthesized multiple chats in the background and updated potentially outdated context. Users could view and edit information and indicate usage preferences through a memory summary. OpenAI’s retrospective stated that the first Dreaming version, referencing conversation history, had been introduced in April 2025.⁠[74]

On July 8, 2026, GPT-Live began powering a new ChatGPT Voice experience. Its full-duplex architecture could receive and produce audio simultaneously and delegate search or complex reasoning to background models while continuing the spoken interaction. GPT-Live-1 and mini variants launched in a staged rollout across web, iOS and Android. Video and screen sharing were initially unsupported, with the retained older voice experience available for those features.⁠[75]

On July 9, 2026, OpenAI introduced ChatGPT Work for delegating longer tasks across files and applications, producing documents, spreadsheets, presentations and websites, and monitoring or redirecting work in progress. It incorporated Codex agent technology and offered a Plugins directory, a Sites beta for hosting interactive websites, and scheduled-task capabilities.⁠[76]

The new desktop app released on July 9, 2026 combined Chat, Work and Codex for users across plans on macOS and Windows. The former Codex app was incorporated into it, and the older ChatGPT desktop app was renamed ChatGPT Classic. OpenAI also announced the beginning of Atlas’s retirement as a standalone browser, with an in-app desktop browser and browser extensions taking over related work.⁠[76]

The GPT-5.6 series officially launched on July 9, 2026 in Sol, Terra and Luna capability and cost tiers. Availability and reasoning levels for models in the series were configured separately across the API, ordinary chat and agent environments.⁠[77]

On July 13, 2026, OpenAI announced broader parental safety notifications for linked teen accounts, including accounts disabled for violent activity. It said notifications concerned situations potentially requiring urgent adult support, not fiction writing, games, news or general political discussion.⁠[56]

On July 23, 2026, OpenAI announced that ChatGPT Health was available on web and iOS to US users aged 18 and older.⁠[64]

On August 6, 2026, OpenAI released country-level usage data and said more than one billion people used ChatGPT, without explicitly defining that number as weekly active users. The Signals analysis came from messages in personal Free, Go, Plus and Pro accounts. Multimedia-related uses accounted for 7.8% of messages, exceeding one in ten in Brazil and Colombia. The proportion using ChatGPT to perform tasks or produce outputs was more than twice as high in work contexts as outside work. These results described usage distributions in the company’s measured sample.⁠[78]

On August 6, 2026, following voluntary withdrawals by news-publisher plaintiffs, the Southern District of New York dismissed certain contributory-infringement claims based on a material-contribution theory concerning end users’ infringement, and denied some applications to add claims. The order did not end the entire copyright dispute.⁠[79]

On August 14, 2026, ChatGPT’s desktop app launched a Linux public preview. Initially, it could operate websites through its built-in browser but could not control other desktop applications.⁠[10]

On August 18, 2026, OpenAI released ChatGPT for Teens, combining Study mode, quizzes, learning visualizations and homework hints in a teen experience. New Study Hours, configurable by teenagers or parents, made Study mode the default during specified times.⁠[80]

Users who reported being 13–17 or were estimated by the system to be under 18 automatically entered the teen experience. Stronger default protections applied to high-risk areas such as self-harm, violence and eating disorders, with restrictions on romanticized language and encouragement of emotional dependence. New prompts also reminded teenagers to avoid uploading sensitive images. These were product-protection designs, not guarantees that every risk would be detected.⁠[80][57]

On September 1, 2026, the US Department of Justice filed a statement of interest supporting fair use at the training stage while distinguishing training from potentially infringing outputs. This legal position was not a court ruling.⁠[81]

On September 3, 2026, OpenAI announced GPT-6 AstraGPT-6GPT-6 is an OpenAI model family introduced in 2026, spanning Astra, Sol, Luna and subsequent 6.1 versions. Its applications include conversation, coding, computer use and professional work, with capabilities and availability varying across models and products.Open the full entry, emphasizing computer operation, file deliverables, coding and longer workflows for complex tasks. Its announcement said availability began that day with a limited set of organizations, with expansion to ChatGPT Plus, Pro, Business and Enterprise users planned over subsequent days. Enterprise workspaces required administrator activation at launch. The announcement did not specify when access would finish rolling out to every account.⁠[82]

On September 4, 2026, OpenAI announced that it had moved for summary judgment. Neither a government legal position nor a party’s motion constituted a final merits judgment by the court.⁠[83]

On November 4, 2026, OpenAI released Images 2.6 across ChatGPT subscription tiers worldwide on desktop, mobile and web. It added GIF creation and improved consistency of characters, products and brand details through image edits. OpenAI reported a 50% improvement in generation speed.⁠[84]

On November 9, 2026, ChatGPT Enterprise became available in AmazonAmazonAmazon is a publicly traded U.S. technology and retail company founded by Jeff Bezos in 1994, with its principal corporate offices in Seattle. Its businesses span e-commerce, logistics and fulfillment, digital advertising, membership and entertainment services, consumer electronics, and Amazon Web Services cloud computing.Open the full entry’s AWSAmazon Web ServicesAmazon Web Services (AWS) is the cloud computing business of Amazon. Launched in July 2002, it made its public debut with the Amazon S3 storage service on March 14, 2006, and has since added hundreds of services, including Amazon EC2; its segment net sales were about $46.0 billion in the third quarter of 2026.Open the full entry Marketplace. Organizations could use their AWS procurement and billing arrangements and count eligible spending toward AWS commitments. The offering retained the models, features and simultaneous updates available directly from OpenAI; eligible customers could also obtain ChatGPT for Healthcare and ChatGPT for Financial Services through the marketplace.⁠[85]

On November 10, OpenAI introduced the Businesses in ChatGPT beta. Verified businesses could manage company details and product pricing and availability through Business Hub, alongside aggregate mentions, clicks and discovery topics. OpenAI said it would not share users’ conversations with businesses. Eligible U.S. ecommerce and online-service companies could launch Business Agents, with distinctly labeled conversations separate from original ChatGPT chats and independent answers. Sponsored placements were also labeled. ShopifyShopifyShopify is a commerce software company that originated in Ottawa, Canada, providing merchants with tools for online stores, physical retail, payments, orders and inventory management. The business began with Snowdevil, an online snowboard store founded in 2004 by Tobias Lütke and others, and the standalone e-commerce platform launched in 2006. The company went public in 2015 and now trades under the ticker SHOP on Nasdaq and the Toronto Stock Exchange.Open the full entry merchants could connect stores to simplify verification.⁠[86]

ChatGPT is an AI service delivered through websites and applications, using conversational context to respond to questions and follow-up instructions. OpenAI’s description of model development identifies three main information sources: publicly available internet material, partner-provided material, and information supplied or generated by users, trainers and researchers. Development involves data preparation, pretraining, post-training and post-deployment evaluations. Models adjust parameters to learn patterns in data and generate text and other outputs incrementally from context; identical questions may receive different answers.⁠[87][88]

The product combines model responses with tool calls. Search retrieves online information and provides sources; file analysis processes documents and spreadsheets; data analysis calculates and creates charts in a code-execution environment; and image and voice features handle other forms of input and output. Availability depends on plan, region, model and staged rollout. Previously released features may subsequently change or be discontinued.⁠[89][90]

Generated content can contain hallucinations: incorrect claims presented as credible facts. In research dated September 5, 2025, OpenAI argued that training and evaluation that reward correct guesses without recognizing expressions of uncertainty encourage models to guess when evidence is insufficient. More reasoning or search does not guarantee accurate answers to every question. Fluency and apparent confidence therefore do not establish factual correctness.⁠[91]

As of September 7, 2026, ChatGPT offered Free, Go, Plus and Pro personal plans, alongside Business, Enterprise and education offerings. These differed in models, tools, usage allowances and administrative features. OpenAI indicated that Go could include advertising. Feature tables and model names could change as the product evolved.⁠[90]

The official help pages consulted that day listed Plus at $20 per month and two Pro tiers at $100 and $200 per month, with allowances respectively five and twenty times Plus usage, subject to feature-specific limits and anti-abuse rules. ChatGPT subscriptions and the OpenAI API were billed separately; Plus did not include API usage. These dollar amounts were official list prices on the access date; applicable regional prices and checkout terms determined actual purchases.⁠[92][93]

OpenAI separately published supported countries and regions for web and mobile access. On September 7, 2026, the list included Taiwan, Singapore, Japan and the United States, among others, but did not list mainland China, Hong Kong or Macau. Geographic service availability was distinct from the ability to converse in Chinese.⁠[94]

Under OpenAI’s data-control documentation, users could choose whether conversations helped improve models. Turning this off still allowed chats to remain in history. Logged-in users could export data or delete accounts, and training preferences synchronized across web and mobile. Training use, conversation history, personalized memory and litigation preservation were separate data-processing matters.⁠[95]

ChatGPT: product web interface. Website: https://chatgpt.com/ (opens in a new window)⁠[96]

Official downloads: desktop, iOS and Android application access. Website: https://chatgpt.com/download/ (opens in a new window)⁠[97]

Plans and pricing: personal and organizational offerings. Website: https://chatgpt.com/pricing/ (opens in a new window)⁠[90]

OpenAI Help Center: account, features, billing and privacy guidance. Website: https://help.openai.com/en (opens in a new window)⁠[98]

Service status: incidents and maintenance affecting ChatGPT and other products. Website: https://status.openai.com/ (opens in a new window)⁠[99]

Privacy requests: OpenAI’s personal-data rights request portal. Website: https://privacy.openai.com/ (opens in a new window)⁠[87]