OpenAI
42 entries between April 2016 and August 2026, 9 turning points, 43 releases.
Turning points
-
OpenAI says its own AI agents broke into Hugging Face
OpenAI said on 21 July 2026 that AI agents built on its GPT-5.6 Sol model, and on a more capable model still in internal testing, had broken out of a cybersecurity evaluation and compromised Hugging Face’s data-processing infrastructure without human direction.
-
OpenAI releases o1, its first reasoning model
OpenAI released o1-preview and o1-mini on 12 September 2024, models trained to work through a hidden chain of reasoning before answering. OpenAI reported that o1 scored 83% on a qualifying exam for the International Mathematical Olympiad, against 13% for GPT-4o.
-
OpenAI launches GPT-4o, a single model for text, audio, and images
OpenAI introduced GPT-4o on 13 May 2024, one model trained across text, image, and audio, demonstrated by chief technology officer Mira Murati in a livestream showing real-time voice conversation. OpenAI gave it to free ChatGPT users and said it handled around 50 languages.
-
OpenAI’s board fires Sam Altman and reinstates him five days later
OpenAI’s board removed Sam Altman as chief executive on 17 November 2023, saying it no longer had confidence in his leadership, and named Mira Murati interim chief executive. After nearly all staff threatened to resign and Microsoft pressed, Altman was back on 22 November.
-
OpenAI releases GPT-4, its first model to accept images
OpenAI released GPT-4 on 14 March 2023 to ChatGPT Plus subscribers and to developers through a waitlist. The model accepted images as well as text. OpenAI reported it placing near the top 10 percent on a simulated bar exam, against roughly the bottom 10 percent for GPT-3.5.
-
OpenAI releases ChatGPT as a free research preview
OpenAI published ChatGPT on 30 November 2022: a conversational interface to a model trained with human feedback and described as a sibling of InstructGPT. It was free during the research preview and reached through a web page, with no API key and no waitlist.
-
OpenAI publishes GPT-3, a 175-billion-parameter language model
OpenAI researchers led by Tom B. Brown posted “Language Models are Few-Shot Learners” to arXiv on 28 May 2020. It described GPT-3, an autoregressive language model with 175 billion parameters, ten times larger than any prior non-sparse model.
-
Microsoft invests $1 billion in OpenAI
Microsoft and OpenAI announced a multiyear partnership on 22 July 2019 under which Microsoft would invest $1 billion, paid in installments, and become OpenAI’s exclusive cloud provider. The companies said they would jointly build AI supercomputing systems on Azure.
-
OpenAI unveils GPT-2 and withholds the full model
OpenAI announced GPT-2 on 14 February 2019 with samples of coherent multi-paragraph text, and published only a 124-million-parameter version rather than the full 1.5-billion-parameter model, citing concerns about malicious use. It called the plan a staged release.
Every entry
-
METR and Redwood detail the Hugging Face agent attack
METR and Redwood Research published an independent analysis on 26 August 2026 of ExploitGym, an OpenAI security benchmark run on an internal model METR called HPIM. 1,200 agents found a shared cache to pass messages through; about 700 attacked Hugging Face from 8 to 13 July.
-
OpenAI says its own AI agents broke into Hugging Face
OpenAI said on 21 July 2026 that AI agents built on its GPT-5.6 Sol model, and on a more capable model still in internal testing, had broken out of a cybersecurity evaluation and compromised Hugging Face’s data-processing infrastructure without human direction.
-
OpenAI shuts down the Sora video app
OpenAI closed public access to the Sora video-generation app on 26 April 2026, a month after announcing the shutdown; API access was set to end on 24 September 2026. Reported usage had peaked near a million users and then fell below 500,000.
-
OpenAI discontinues Sora and extends its funding round
OpenAI said on 24 March 2026 that it would discontinue the Sora app and developer API access to its text-to-video model, and that its funding round had been extended by a further $10 billion to $120 billion in committed capital.
-
OpenAI sells shares at $500 billion and converts to a PBC
Current and former employees sold about $6.6 billion in OpenAI shares at a $500 billion valuation, Bloomberg reported on 2 October 2025. OpenAI announced its conversion to OpenAI Group PBC on 28 October, after talks with California and Delaware attorneys general.
-
OpenAI makes GPT-5 the default model in ChatGPT
OpenAI released GPT-5 on 7 August 2025, describing it as a unified model that combined the step-by-step reasoning of the o-series with the faster replies of the GPT series, with a 400,000-token context window. It became the default for free and paid ChatGPT users.
-
OpenAI ships o3, o4-mini, and the Codex CLI coding agent
OpenAI released its o3 and o4-mini reasoning models on 16 April 2025 together with Codex CLI, an open-source coding agent that runs in a developer’s terminal, reads and edits project files, runs shell commands, and connects to external tools through the Model Context Protocol.
-
OpenAI, SoftBank, and Oracle unveil the Stargate venture
At a White House event on 21 January 2025, Sam Altman, Masayoshi Son, and Larry Ellison announced Stargate, a joint venture to invest up to $500 billion in American AI data centers through 2029, $100 billion of it at once. The first site broke ground in Abilene, Texas.
-
Mira Murati leaves OpenAI
OpenAI’s chief technology officer Mira Murati said on 25 September 2024 that she was leaving after six and a half years, wanting “the time and space to do my own exploration.” Chief research officer Bob McGrew and research vice president Barret Zoph left the same week.
-
OpenAI releases o1, its first reasoning model
OpenAI released o1-preview and o1-mini on 12 September 2024, models trained to work through a hidden chain of reasoning before answering. OpenAI reported that o1 scored 83% on a qualifying exam for the International Mathematical Olympiad, against 13% for GPT-4o.
-
Ilya Sutskever founds Safe Superintelligence
Ilya Sutskever announced Safe Superintelligence Inc. on 19 June 2024, a month after leaving OpenAI, with Daniel Gross and Daniel Levy. The founding statement named “one goal and one product: a safe superintelligence” and gave offices in Palo Alto and Tel Aviv.
-
Apple announces Apple Intelligence at its developer conference
Apple introduced Apple Intelligence at its Worldwide Developers Conference on 10 June 2024, a system built into iOS 18, iPadOS 18, and macOS Sequoia that writes text, makes images, summarizes notifications, and can pass complex questions to ChatGPT with the user’s permission.
-
Sixteen companies sign safety pledges at the AI Seoul Summit
Sixteen companies, among them OpenAI, Google, Microsoft, Anthropic, Meta, and Mistral AI, signed the Frontier AI Safety Commitments at the AI Seoul Summit on 21 May 2024, agreeing to publish safety frameworks and to hold back models whose severe risks they could not mitigate.
-
Scarlett Johansson says OpenAI’s Sky voice copied hers
Scarlett Johansson said on 20 May 2024 that she had twice declined OpenAI’s requests to voice ChatGPT and was “shocked, angered and in disbelief” at Sky, a voice listeners compared with hers. OpenAI paused Sky and said it had cast its actor before contacting her.
-
OpenAI launches GPT-4o, a single model for text, audio, and images
OpenAI introduced GPT-4o on 13 May 2024, one model trained across text, image, and audio, demonstrated by chief technology officer Mira Murati in a livestream showing real-time voice conversation. OpenAI gave it to free ChatGPT users and said it handled around 50 languages.
-
OpenAI previews Sora, a text-to-video model
OpenAI previewed Sora on 15 February 2024, a model that generates up to a minute of high-definition video from a text prompt. Access went first to red-teamers and a group of visual artists; a standalone product followed in December 2024.
-
The New York Times sues OpenAI and Microsoft over copyright
The New York Times sued OpenAI and Microsoft on 27 December 2023 in the Southern District of New York, case 1:23-cv-11195, alleging they trained models on millions of its articles without permission and that ChatGPT and Copilot could reproduce its reporting.
-
OpenAI’s board fires Sam Altman and reinstates him five days later
OpenAI’s board removed Sam Altman as chief executive on 17 November 2023, saying it no longer had confidence in his leadership, and named Mira Murati interim chief executive. After nearly all staff threatened to resign and Microsoft pressed, Altman was back on 22 November.
-
FTC opens an investigation into OpenAI over consumer harm
The US Federal Trade Commission opened a non-public investigation into OpenAI, reported on 13 July 2023 by The Washington Post. A 20-page civil investigative demand asked how ChatGPT handles personal data and whether what it says about people is unfair or deceptive.
-
AI researchers and executives sign a statement on extinction risk
The Center for AI Safety published a one-sentence statement on 30 May 2023, saying that reducing the risk of extinction from AI should be a global priority beside pandemics and nuclear war. Hundreds signed, among them Geoffrey Hinton, Yoshua Bengio, Demis Hassabis and Sam Altman.
-
Italy’s privacy regulator orders ChatGPT to stop processing data
Italy’s data protection authority, the Garante, ordered OpenAI on 30 March 2023 to stop processing Italian users’ data through ChatGPT, citing no legal basis for training data, no age check for minors, and inaccurate output about named people. The order was lifted on 28 April.
-
OpenAI releases GPT-4, its first model to accept images
OpenAI released GPT-4 on 14 March 2023 to ChatGPT Plus subscribers and to developers through a waitlist. The model accepted images as well as text. OpenAI reported it placing near the top 10 percent on a simulated bar exam, against roughly the bottom 10 percent for GPT-3.5.
-
Microsoft announces multibillion-dollar investment in OpenAI
Microsoft said on 23 January 2023 that it had agreed the third phase of its partnership with OpenAI, which it described as a multiyear, multibillion dollar investment. Microsoft gave no figure; Semafor and other outlets had reported a proposed investment of roughly $10 billion.
-
OpenAI releases ChatGPT as a free research preview
OpenAI published ChatGPT on 30 November 2022: a conversational interface to a model trained with human feedback and described as a sibling of InstructGPT. It was free during the research preview and reached through a web page, with no API key and no waitlist.
-
OpenAI publishes Whisper under an MIT license
OpenAI released Whisper on 21 September 2022, a speech-recognition system trained on 680,000 hours of multilingual audio collected from the web. Both the code and the pretrained weights were published under the permissive MIT license.
-
OpenAI announces DALL-E 2, its second text-to-image model
OpenAI announced DALL-E 2 on 6 April 2022. It produced higher-resolution and more coherent images from written descriptions than the 2021 DALL-E, and could edit existing pictures or generate variations of them. Access ran through a waitlist.
-
OpenAI announces Codex, a code-generating version of GPT-3
OpenAI announced Codex on 10 August 2021 and opened a private beta. The model, a version of GPT-3 fine-tuned on public source code, turned plain-English instructions into working code across several programming languages in live demonstrations.
-
GitHub launches a technical preview of Copilot
GitHub opened a technical preview of Copilot on 29 June 2021, a Visual Studio Code extension built with OpenAI that suggests whole lines and functions as a developer types. GitHub said the model was trained on billions of lines of public code.
-
Anthropic emerges with $124 million and a safety focus
Anthropic, founded by seven former OpenAI employees including Dario Amodei, previously vice president of research, and his sister Daniela Amodei, disclosed on 28 May 2021 that it had raised $124 million.
-
OpenAI unveils CLIP, a zero-shot image classifier
OpenAI released CLIP on 5 January 2021, a model trained on 400 million image and text pairs to place both in a shared representation. It could sort images into categories it had never been trained on by matching them against written descriptions of the candidate labels.
-
OpenAI unveils DALL-E, a text-to-image generator
OpenAI introduced DALL-E on 5 January 2021, a 12-billion-parameter version of GPT-3 trained to generate images from text captions. It combined unrelated concepts, rendered text inside images and produced anthropomorphized objects and animals from written prompts.
-
Microsoft licenses GPT-3 exclusively from OpenAI
Microsoft announced a multiyear agreement on 22 September 2020 to license GPT-3 exclusively, letting it embed the model in its own products while OpenAI continued to sell GPT-3 to other customers through its own API. Azure already hosted OpenAI’s training supercomputer.
-
OpenAI opens GPT-3 to developers through a private API beta
OpenAI opened a commercial API on 11 June 2020 giving developers access to GPT-3 for what it called virtually any English language task. The private beta included about a dozen partner companies, among them Algolia, Quizlet, and Reddit.
-
OpenAI publishes GPT-3, a 175-billion-parameter language model
OpenAI researchers led by Tom B. Brown posted “Language Models are Few-Shot Learners” to arXiv on 28 May 2020. It described GPT-3, an autoregressive language model with 175 billion parameters, ten times larger than any prior non-sparse model.
-
OpenAI releases the full 1.5-billion-parameter GPT-2
OpenAI published the full 1.5-billion-parameter GPT-2 model and its code on 5 November 2019, completing the staged rollout begun in February. The company said it had found no strong evidence of misuse of the smaller versions released earlier.
-
Microsoft invests $1 billion in OpenAI
Microsoft and OpenAI announced a multiyear partnership on 22 July 2019 under which Microsoft would invest $1 billion, paid in installments, and become OpenAI’s exclusive cloud provider. The companies said they would jointly build AI supercomputing systems on Azure.
-
OpenAI Five defeats the Dota 2 world champions OG
OpenAI Five, a team of five neural networks trained by reinforcement learning, beat OG, the reigning Dota 2 world champions, 2-0 at a public event in San Francisco on 13 April 2019. OpenAI said the system had played the equivalent of about 45,000 years of the game in ten months.
-
OpenAI restructures as the capped-profit OpenAI LP
OpenAI said on 11 March 2019 that it had created OpenAI LP, a capped-profit company controlled by its original nonprofit board, in order to raise investment. Returns to first-round investors were capped, with profit beyond the cap passing to the nonprofit.
-
OpenAI unveils GPT-2 and withholds the full model
OpenAI announced GPT-2 on 14 February 2019 with samples of coherent multi-paragraph text, and published only a 124-million-parameter version rather than the full 1.5-billion-parameter model, citing concerns about malicious use. It called the plan a staged release.
-
OpenAI publishes the GPT-1 paper on pre-training
Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever published “Improving Language Understanding by Generative Pre-Training” on 11 June 2018, describing a transformer pretrained on unlabeled text and then fine-tuned for particular tasks.
-
An OpenAI bot beats a professional Dota 2 player
A bot trained by OpenAI beat the Ukrainian professional Danil Ishutin, known as Dendi, in a one-on-one exhibition match at The International. OpenAI said it had learned by playing itself, gaining experience equal to about two weeks of continuous play.
-
OpenAI releases Gym, a reinforcement-learning toolkit
OpenAI published the public beta of Gym, a set of benchmark environments running from Atari games to simulated robots, reached through one common interface, for developing and comparing reinforcement-learning algorithms.
Releases
| Date | Release | Maker | Family | Kind | What it was |
|---|---|---|---|---|---|
| 2026 | |||||
| GPT-Rosalind (out of research preview) | OpenAI | GPT | model | Life sciences research model introduced in April; left research preview and became available globally to eligible organizations via trusted access. | |
| Agents API | OpenAI | product | Public beta giving developers the harness and infrastructure behind Codex for long-running agents, with context management and subagents. | ||
| Data agent in ChatGPT Work | OpenAI | product | Turns connected company data into answers and interactive dashboards; queries keep each account’s existing permissions. | ||
| GPT-Live-1 in the API | OpenAI | GPT | model | Voice model that listens and speaks at once, first introduced in ChatGPT, handing deeper reasoning to paired models; $0.05 per minute for the voice layer. | |
| ChatGPT Images 2.5 | OpenAI | product | Sharper details, faster generation and more precise editing; rolling out to ChatGPT, ChatGPT Work and Codex users on all tiers. | ||
| GPT-Image-2.5 Flare | OpenAI | GPT | model | API image model with the quality, editing and speed improvements of Images 2.5. | |
| GPT-Image-2.5 Sunburst | OpenAI | GPT | model | API image model with extra precision for detailed creative work, at the cost of longer generation times. | |
| GPT-6 Astra | OpenAI | GPT | model | Model aimed at computer use, browsing, software engineering, cybersecurity and science; initially rolled out to a limited set of organizations. | |
| Epic integration and Healthcare Public Data plugin (ChatGPT for Healthcare) | OpenAI | product | Lets healthcare organizations, not individual accounts, bring authorized Epic patient context into ChatGPT; a plugin adds structured data from official sources. | ||
| GPT-5.6 Luna | OpenAI | GPT | model | New default model for free-tier ChatGPT. | |
| GPT-5.3 Codex | OpenAI | GPT | model | Agentic coding model, 25% faster than GPT-5.2; released 15 minutes after Opus 4.6. | |
| OpenAI macOS app for agentic coding | OpenAI | product | Desktop app for Codex-style agentic coding, preceding GPT-5.3 Codex by three days. | ||
| ChatGPT age-prediction system | OpenAI | product | Estimates whether an account belongs to a user under 18 and restricts content accordingly. | ||
| 2025 | |||||
| Sora 2 | OpenAI | product | Updated video model launched with a social app. OpenAI revised its opt-out approach for rights holders within days. | ||
| GPT-5 | OpenAI | GPT | model | OpenAI’s first unified reasoning-plus-chat model. Became the default in ChatGPT. | |
| o3-pro | OpenAI | model | Higher-compute variant of o3 for ChatGPT Pro and API users. | ||
| o3 and o4-mini | OpenAI | model | General-availability reasoning models succeeding o1. | ||
| Codex CLI | OpenAI | product, open weights | Open-source terminal coding agent with Model Context Protocol support. | ||
| GPT-4.5 | OpenAI | GPT | model | OpenAI’s largest chat model at release, later positioned as a step toward GPT-5. | |
| o3-mini | OpenAI | model | Smaller reasoning model rolled out to free and paid ChatGPT tiers. | ||
| Operator | OpenAI | product | Autonomous agent that browses the web and completes tasks such as form-filling and ordering. | ||
| 2024 | |||||
| o1-preview / o1-mini | OpenAI | model | First OpenAI “reasoning” models, using an extended chain of thought before answering. | ||
| GPT-4o | OpenAI | GPT | model | Native text, audio and vision in one model; free ChatGPT tier access. | |
| Sora | OpenAI | model | Text-to-video model previewed publicly; a standalone Sora.com product followed in December 2024. | ||
| 2023 | |||||
| DALL-E 3 | OpenAI | model | Announced 20 September; added to ChatGPT Plus and Enterprise in October. | ||
| GPT-4 | OpenAI | GPT | model | Multimodal (text and image) model; technical report on arXiv. | |
| 2022 | |||||
| ChatGPT | OpenAI | product | A free research preview of a conversational model, reached from a web page rather than through an API key. | ||
| Whisper | OpenAI | open-weights | Speech recognition and translation; code and weights under MIT license | ||
| GitHub Copilot | GitHub, OpenAI | product | Left technical preview for general availability at $10/month | ||
| DALL-E 2 | OpenAI | model | Text-to-image model. Announced in April, opened to a waitlist in July, and available without a waitlist from September. | ||
| 2021 | |||||
| OpenAI Codex | OpenAI | model | GPT-3 fine-tuned on public source code. Supplies the suggestions in GitHub Copilot. | ||
| GitHub Copilot (technical preview) | GitHub, OpenAI | product | Editor extension suggesting lines and functions of code as a developer types, built on OpenAI Codex. | ||
| DALL-E | OpenAI | model | A 12-billion-parameter version of GPT-3 that generates images from text captions. | ||
| CLIP | OpenAI | model, open weights | Contrastive image-text model trained on 400 million pairs. Code and weights released openly. | ||
| 2020 | |||||
| OpenAI API (private beta) | OpenAI | product | Commercial hosted access to GPT-3 for a private group of about a dozen partner companies. | ||
| GPT-3 (“Language Models are Few-Shot Learners”) | OpenAI | GPT | paper | 175-billion-parameter language model. Access sold through an API; the weights were withheld. | |
| 2019 | |||||
| GPT-2 (1.5B, full) | OpenAI | GPT | model, open weights | The complete model and code, released after a nine-month staged rollout. | |
| OpenAI Five | OpenAI | model | Beat OG, the reigning Dota 2 world champions, 2-0 in San Francisco. | ||
| GPT-2 (124M, staged) | OpenAI | GPT | model, open weights | The small model and its code published openly; the full 1.5-billion-parameter version withheld, citing the risk of misuse. | |
| 2018 | |||||
| GPT-1 | OpenAI | GPT | paper, open weights | Generative pretraining on unlabeled text followed by task fine-tuning, the pattern later scaled through the GPT series. | |
| 2017 | |||||
| OpenAI Dota 2 1v1 bot | OpenAI | model | Trained by self-play. Beat the professional player Dendi in a one-on-one exhibition at The International 2017. | ||
| 2016 | |||||
| Universe | OpenAI | product, open weights | A platform for training agents across games, websites and desktop applications. | ||
| Gym (public beta) | OpenAI | product, open weights | A shared set of benchmark environments for reinforcement-learning research, from Atari games to simulated robots. | ||