This evergreen history article uses authoritative archives and official records. Exact dates are used when documented; gradual inventions and rollouts are described as periods rather than being assigned a misleading single birthday.
Quick facts
- OpenAI was announced in December 2015 in San Francisco, California.
- The organisation began with a mission focused on broadly beneficial artificial intelligence research.
- The GPT series applied transformer-based pretraining to general language tasks.
- ChatGPT was released publicly in November 2022 and made conversational large language models accessible to a mass audience.
- OpenAI’s history includes research, commercial products, safety debates, partnerships and major governance changes.
Founding and the research-lab period
OpenAI was introduced in December 2015 by a group of technology entrepreneurs and researchers including Sam Altman, Elon Musk, Greg Brockman, Ilya Sutskever, Wojciech Zaremba and John Schulman. The organisation was established in San Francisco as a nonprofit research laboratory. Its public mission was to help ensure that advanced artificial intelligence benefited humanity broadly.
Early work included reinforcement learning, robotics, game-playing environments and open research tools. OpenAI Gym, released in 2016, provided standard environments for comparing reinforcement-learning algorithms. The organisation also explored multi-agent learning and systems that mastered complex games through large-scale self-play.
From nonprofit laboratory to capped-profit structure
Training increasingly large models required substantial computing resources and engineering investment. In 2019 OpenAI created a capped-profit structure intended to attract capital while retaining mission-based oversight through the nonprofit. Microsoft announced a major partnership and became a key cloud-computing provider.
The structure generated continuing debate. Supporters argued that frontier research required industrial-scale funding; critics questioned transparency, concentration of power and whether commercial incentives could remain aligned with a public-interest mission. Governance later became one of the most closely watched parts of OpenAI’s story.
The GPT series and general-purpose language models
The first Generative Pre-trained Transformer model appeared in 2018. The approach trained a model on broad text before adapting it to specific tasks. GPT-2 in 2019 demonstrated stronger coherent generation, while GPT-3 in 2020 scaled the method dramatically and showed that a single model could perform many tasks from prompts and examples.
This model family shifted attention from narrowly trained language systems toward general interfaces based on natural-language instructions. The same research direction led to code generation through Codex and image generation through DALL·E. OpenAI increasingly offered models through cloud APIs, allowing outside developers to build products without training comparable systems themselves.
ChatGPT changes public awareness
ChatGPT launched as a research preview on 30 November 2022. It combined a conversational interface with a language model refined using human feedback. People could ask follow-up questions, request revisions, generate drafts and interact without learning a programming interface. Adoption spread rapidly through schools, workplaces, creative industries and software development.
The launch also exposed limitations at unprecedented scale. ChatGPT could produce confident false statements, reproduce bias, assist harmful requests or generate content without clear provenance. Education systems debated assessment, publishers considered attribution and copyright, and governments began drafting rules for generative AI.
Multimodal models, tools and the agentic direction
GPT-4, released in 2023, expanded capability and accepted image input in supported settings. Later product development connected models to web search, data analysis, voice, image generation and external tools. ChatGPT evolved from a text chatbot into a broader workspace for research, coding, media creation and task completion.
OpenAI’s future is tied to model reliability, infrastructure cost, competition, safety evaluation and governance. The company’s impact reaches beyond any single model: it helped establish conversational AI as a general computing interface and forced businesses, educators and policymakers to confront the opportunities and risks of widely accessible generative systems.
Common misconceptions
- OpenAI did not invent the transformer architecture; it built influential model families using and extending that research direction.
- ChatGPT is a product and interface, while GPT refers to a family of underlying model architectures and releases.
- A fluent response is not proof that the model has verified every claim.
- OpenAI has not remained organisationally unchanged since 2015; its structure, partnerships and governance have evolved.
Timeline: key years and locations
| Year | Location | Event | Why it mattered |
|---|---|---|---|
| 2015 | San Francisco, California, United States | OpenAI is announced as a nonprofit research organisation | Created a high-profile laboratory focused on advanced AI. |
| 2016 | San Francisco / global research community | OpenAI Gym is released | Provided common environments for reinforcement-learning research. |
| 2018 | San Francisco, United States | The first GPT paper is published | Demonstrated general language pretraining followed by task adaptation. |
| 2019 | San Francisco and Redmond, Washington | Capped-profit structure and Microsoft partnership develop | Supplied capital and cloud infrastructure for larger models. |
| 2020 | United States | GPT-3 and the OpenAI API expand access | Enabled developers to use a large general-purpose language model. |
| 2021 | United States | DALL·E and Codex are introduced | Extended generative models into images and programming. |
| 2022 | Global online release | ChatGPT research preview launches | Brought conversational generative AI to a mass audience. |
| 2023 onward | Global | Multimodal models and tool-connected assistants expand | Moved the product toward a general AI workspace and agent platform. |
Frequently asked questions
When was ChatGPT launched?
OpenAI released ChatGPT publicly as a research preview on 30 November 2022.
Where was OpenAI founded?
The organisation was established in San Francisco, California, in 2015.
Why was ChatGPT historically important?
It placed a powerful language model behind a simple conversational interface and accelerated global adoption, competition and policy debate.