On 30 November 2022 OpenAI put ChatGPT on the open web. This thread follows what can be dated after that: the real GPT and o-series models OpenAI announced, how quickly people used the product, how much generated text and video then appeared, and how platforms began to label it. Intelligence Age Gulf wrote this page on 30 September 2026 and updated it on 22 October 2026. It does not claim we were publishing in 2022. Model names are the official ones.
ChatGPT: from a free research preview to a web full of generated text
30 November 2022
OpenAI releases ChatGPT as a free research preview
OpenAI published ChatGPT as a conversational research preview, free to try, and asked the public for feedback on what it did well and where it failed. That is the start of this story. The page is dated 30 November 2022.
- Primary source: OpenAI — Introducing ChatGPT · Archived copy
1 January 2023 (month)
UBS estimated 100 million monthly users in January 2023
On 1 February 2023 Reuters reported a UBS note that estimated ChatGPT had reached 100 million monthly active users in January, about two months after launch, citing Similarweb traffic. That is an analyst estimate, not an OpenAI census. Reuters also quoted UBS saying it could not recall a faster ramp in a consumer internet app.
14 March 2023
OpenAI releases GPT-4
On 14 March 2023 OpenAI published GPT-4, a large multimodal model that takes image and text inputs and writes text. It said GPT-4 scored around the top 10% of test takers on a simulated bar exam, while GPT-3.5 scored around the bottom 10%. ChatGPT Plus subscribers could use the text model that day. It was the first major model after the ChatGPT preview.
- Primary source: OpenAI — GPT-4
19 September 2023
TikTok adds a tool for creators to label AI-generated content
TikTok said it would launch a tool so creators could mark videos that were generated or heavily edited by AI, and that it would start testing automatic labels. TechCrunch reported the announcement on 19 September 2023. TikTok already required labels on realistic synthetic media. How much of the app is generated, and how well the labels work, is not something we have a sourced count for.
14 November 2023
YouTube says creators must disclose realistic AI or synthetic video
YouTube said it would require creators to disclose when they upload realistic altered or synthetic content, including work made with AI tools, and that it would show labels to viewers. AP reported the same day. The company said creators who repeatedly did not disclose could face content removal, suspension from the YouTube Partner Program or other penalties. The upload tool itself followed in 2024.
6 February 2024
Meta says it will label AI-generated images on Facebook, Instagram and Threads
Meta said it already marked photorealistic pictures from its own Meta AI feature as “Imagined with AI”, and that in the coming months it would label images uploaded by people when it could detect industry-standard AI markers such as C2PA and IPTC. It also said it was not yet possible to identify all AI-generated content.
18 March 2024
YouTube turns the disclosure rule into a Creator Studio tool
YouTube said creators now had to tell viewers when realistic content — something a person could mistake for a real person, place, scene or event — was made with altered or synthetic media, including generative AI. Labels would appear in the description, and more prominently on sensitive topics. The company said it would not require disclosure for clearly unrealistic work or for using AI only to draft a script.
5 April 2024
Meta says “Made with AI” labels will cover more video, audio and images from May 2024
Meta said it would start adding “Made with AI” labels to a wider range of video, audio and images in May 2024, when it detected industry-standard indicators or when people said they were uploading generated content. It said it agreed with its Oversight Board’s recommendation of a “less restrictive” approach to manipulated media, such as labels with context. On 1 July 2024 Meta added an update to the same note: it was changing the “Made with AI” label to “AI info”.
9 May 2024
TikTok says it will auto-label uploads that already carry C2PA credentials
AP reported that TikTok would begin labelling AI-generated content when it was uploaded from outside the app and already carried Content Credentials from the Coalition for Content Provenance and Authenticity. The company said it already labelled effects made inside TikTok and had required creators to label realistic generated media for more than a year.
- Primary source: Associated Press — TikTok to start labeling AI-generated content
13 May 2024
OpenAI introduces GPT-4o and puts more of it on free ChatGPT
On 13 May 2024 OpenAI introduced GPT-4o as its newest flagship: GPT-4-level intelligence that was faster and better across text, voice and vision. It began rolling the model out to Plus and Team users, and to free ChatGPT with usage limits. It said more than a hundred million people used ChatGPT every week. A later, cheaper GPT-4o mini is listed in OpenAI’s model notes; we have not opened a standalone launch post with a day we trust, so that name is not given its own step.
- Primary source: OpenAI — Introducing GPT-4o and more tools to ChatGPT free users
19 May 2024
The press starts calling the junk “slop”
The Guardian described a wave of low-quality AI pages as “slop” and linked it to a “zombie internet” of generated text and images. Later that summer The Atlantic used the same word for high-volume, low-effort generated posts on social platforms. The word is informal. The useful fact is simpler: after ChatGPT, a lot of cheap generated material appeared in search and on feeds, and people needed a name for the worst of it.
- Primary source: The Guardian — Spam, junk … slop? The latest wave of AI behind the zombie internet
- Primary source: The Atlantic — The MAGA Aesthetic Is AI Slop
12 September 2024
OpenAI starts the o1 reasoning series
On 12 September 2024 OpenAI released o1-preview and o1-mini in ChatGPT and the API. These models were trained to spend more time thinking before they answer. OpenAI said that, on a qualifying exam for the International Mathematics Olympiad, GPT-4o solved 13% of problems and the new reasoning model scored 83%. It reset the series counter to 1 and named it OpenAI o1.
- Primary source: OpenAI — Introducing OpenAI o1-preview
17 September 2024
Google describes its C2PA work and SynthID watermarks
Google said it had joined the C2PA steering committee earlier in 2024 and was putting C2PA Content Credentials and its SynthID watermark into more of its own generative tools. The aim, it said, was to help people see how a picture, film or sound had been made or changed as it moved across the web.
27 February 2025
OpenAI previews GPT-4.5
On 27 February 2025 OpenAI released a research preview of GPT-4.5, which it called its largest and best GPT model for chat so far. It described two complementary ways of scaling: unsupervised learning (GPT-3.5, GPT-4, GPT-4.5) and reasoning (o1, later o3). GPT-4.5 does not think before it responds. Pro users could pick it that day; Plus and Team were due the following week.
- Primary source: OpenAI — Introducing GPT-4.5
14 April 2025
OpenAI launches GPT-4.1 in the API
On 14 April 2025 OpenAI launched GPT-4.1, GPT-4.1 mini and GPT-4.1 nano in the API. It said they outperformed GPT-4o on coding and instruction following, with context windows up to a million tokens. It also said it would turn off GPT-4.5 Preview in the API on 14 July 2025. This was an API launch, not a new consumer ChatGPT generation.
- Primary source: OpenAI — Introducing GPT-4.1 in the API
16 April 2025
OpenAI releases o3 and o4-mini
On 16 April 2025 OpenAI published the o3 and o4-mini system card. Those models combine reasoning with tools such as browsing, Python and image analysis. The English “introducing” page did not load when we opened it; the date and names are taken from the system card (Tier 1) and from OpenAI’s Chinese introduction of the same launch. On 28 May 2026 OpenAI’s Help Center said o3 would be retired from ChatGPT on 26 August 2026, with no change to the API.
- Primary source: OpenAI — OpenAI o3 and o4-mini System Card
- Primary source: OpenAI Help Center — Model Release Notes — Retiring OpenAI o3 and GPT-4.5 · Archived copy
7 August 2025
OpenAI introduces GPT-5
On 7 August 2025 OpenAI introduced GPT-5 as a unified system: a faster model for most questions, a deeper “thinking” model for harder ones, and a router that chooses between them. It said the system was available to all ChatGPT users, with more usage for Plus and a GPT-5 pro mode for Pro subscribers.
- Primary source: OpenAI — Introducing GPT-5
25 September 2025
The National reports a UAE Media Council warning on AI depictions of public figures
The National reported that the UAE Media Council warned against using AI to depict public figures and national symbols without prior official approval, and against using it to spread misinformation. We have that report. We have not opened a Media Council original, so this step is attributed to The National, not to a gazette text.
19 May 2026
Google says Search and Chrome will check SynthID and C2PA
At Google I/O 2026, Google said people would be able to check images for SynthID watermarks and C2PA Content Credentials in the Gemini app that day, and later in Search and Chrome. The Verge reported the same announcement. That is a detection and provenance step, not a claim that every generated picture on the web can now be spotted.
26 June 2026
OpenAI previews GPT-5.6 Sol, Terra and Luna
On 26 June 2026 OpenAI began a limited preview of the GPT-5.6 series: Sol as the flagship, Terra as a balanced everyday model, and Luna as a cheaper one. Broader release followed on 9 July 2026. Sol is the model later named, with a more capable pre-release sibling, in OpenAI’s account of the July 2026 Hugging Face evaluation incident.
- Primary source: OpenAI — Previewing GPT-5.6 Sol
9 July 2026
GPT-5.6 Sol becomes generally available
On 9 July 2026 OpenAI said GPT-5.6 Sol, Terra and Luna were generally available in ChatGPT, Codex and the API. The Help Center recorded the ChatGPT rollout the same day for eligible paid plans. This is the official launch day. It is also the first day of the window Hugging Face later reconstructed for the evaluation-related intrusion.