OpenAI Models & Timeline

OpenAI was established as a nonprofit research organization with the goal of advancing safe and beneficial artificial intelligence.

OpenAI released GPT-2, demonstrating impressive language-generation capabilities (initially held back for safety considerations).

The release of GPT-3 marked a significant breakthrough, powering numerous applications and laying the groundwork for later products.

ChatGPT was introduced as a free research preview. Its conversational abilities and ease of use quickly drove massive public interest and adoption, setting the stage for rapid iterative improvements.

Over the following months and years, OpenAI refined its offerings, moved to a “capped-profit” model, and expanded features (including subscription plans, enterprise solutions, and customizability) that have reshaped both the user experience and the underlying business model.

OpenAI announced experimental support for AI plugins, enabling ChatGPT to use tools such as browsing, code execution (code interpreter), and third-party integrations.

Users gained the ability to turn off chat history and export data, and OpenAI announced the deprecation of the legacy GPT‑3.5 model (with new messages defaulting to a more advanced model).

A new beta panel enabled users to try out enhanced web browsing and plugin features directly from ChatGPT’s settings.

The iOS app expanded its reach into more countries, introduced shared links (in alpha), integrated a Bing Plugin, and added the option to disable chat history.

Updates improved the mobile ChatGPT experience, refining how browsing works and making search results jump directly to the relevant chat context.

Due to issues with content display, the browsing beta was temporarily disabled while fixes were implemented.

ChatGPT Plus users began receiving access to the code interpreter beta, allowing the assistant to execute Python code, analyze data, generate charts, and more.

The message limit for Plus customers using GPT‑4 was doubled (to 40 every 3 hours), allowing for longer or more complex interactions.

A beta rollout of custom instructions gave users greater control over how ChatGPT responds by letting them set preferences that persist across conversations.

The dedicated Android app was introduced in key markets (such as the United States, India, Bangladesh, and Brazil) with plans for broader expansion.

A series of small improvements were introduced: new prompt examples, suggested replies, automatic retention of the selected model (with GPT‑4 as default for Plus users), multiple file upload capabilities, enhanced login persistence, and keyboard shortcuts.

Custom instructions became available to free users (with a phased rollout in the EU & UK on August 21), further allowing personalization of the ChatGPT experience.

A new enterprise plan was launched featuring enhanced security, unlimited faster GPT‑4 access, longer context windows, advanced data analysis, and dedicated workspaces.

The interface began offering a limited set of language options (e.g., Chinese, French, German, Japanese) in an opt‑in alpha for web users.

New voice features (beta on iOS and Android) and general availability of image input for Plus users were rolled out, allowing natural spoken conversations and visual interactions.

After previous adjustments, the browsing feature was rolled back out to Plus users, allowing up‑to‑date research and source linking.

DALL·E 3 was integrated into ChatGPT in beta, enabling users to generate images directly from textual prompts on both web and mobile platforms.

The browsing feature was fully rolled out (no longer in beta), improving the accessibility of current information for Plus and Enterprise users.

Custom versions of ChatGPT (“GPTs”) were introduced, allowing users to create tailored assistants for specific tasks, with the eventual launch of a GPT Store for discovery and monetization.

Voice functionality was expanded to all free users, making it easy to initiate voice conversations via the mobile apps.

OpenAI unveiled a new GPT Store to highlight custom GPTs alongside a team plan aimed at collaborative work with enhanced data privacy and administrative controls.

ChatGPT began testing a “memory” feature that allows the assistant to retain details across sessions, reducing the need to repeat information, with full on/off controls.

ChatGPT was made instantly accessible without the need for registration, lowering the barrier for new users to try the technology.

The memory feature was extended to all ChatGPT Plus subscribers (with some regional rollouts pending), further personalizing user interactions.

A new flagship model—GPT‑4o—was launched, promising GPT‑4–level intelligence with faster performance and improved text, voice, and vision capabilities.

Improvements were made to data analysis tools, including direct file uploads from cloud storage, interactive tables/charts, and presentation-ready outputs, powered by GPT‑4o.

A cost‑efficient, smaller model—GPT‑4o mini—was introduced, offering strong performance on academic benchmarks and enhanced multimodal reasoning.

Free users of ChatGPT were granted limited (up to two images per day) access to DALL·E 3, expanding creative capabilities for non‑subscribers.

Advanced voice mode was introduced on the web (complementing mobile/desktop apps) to facilitate natural, real‑time voice interactions for all paid users.

A comprehensive redesign of the ChatGPT web interface was launched, including a revamped sidebar, improved mobile web experience, and various performance enhancements.

The Canvas feature was introduced, allowing users to edit full model responses graphically, execute Python code within a canvas, and interact with a new “toolbox” integration—especially within GPT‑4o.

Fun and functional updates included a Santa voice option (with a one‑time advanced voice limit reset) plus the rollout of video and screen share capabilities in voice mode (primarily on mobile).

ChatGPT Projects were launched for Plus, Team, and Pro users, enabling grouping of chats and files, sharing context across conversations, and integrated tools like Canvas, advanced data analysis, and DALL·E.

A beta rollout introduced scheduled tasks, allowing users to set reminders or recurring actions that ChatGPT executes automatically at specified future times.

Users can now import entire chat conversations directly into Canvas for further editing, including capabilities to modify code blocks and model responses.

The custom instructions feature received a major update, featuring a new UI that makes it easier to specify desired traits, tone, and rules for ChatGPT’s responses.

Introduction of a companion window for side‑by‑side access, enhanced support for data analysis (including interactive tables/charts), new keyboard shortcuts, and several customization options.

Addition of advanced voice mode (supporting hands‑free interaction), and early support for new models like OpenAI o1‑preview and o1‑mini. Also featured was a refined text fade animation and various performance improvements.

The chat bar was redesigned with an integrated model picker, and the app received further fixes (e.g., pasting from Microsoft Office, restoring keyboard shortcuts).

Introduction of slash commands for quick actions, plus extended integration with popular coding and development applications (such as various VS Code forks and JetBrains IDEs).

New features enabled direct interaction with other apps (like note‑taking and coding apps), added conversation search functionality, and supported better integration with tools like Apple Notes, Notion, and more.

The macOS app was updated to support creating and editing Canvas sessions using GPT‑4o, along with several fixes (such as improved handling of image attachments and conversation renaming).