ChatGPT was released to the public on November 30, 2022, as a free research preview built on the GPT-3.5 language model. OpenAI announced it in a blog post that day, and the product went live at chat.openai.com within hours. It reached roughly one million users in five days, an adoption curve that reset expectations for consumer software. The release date matters because nearly every later capability, policy fight, and competitor launch is dated relative to that Wednesday.

The mechanism: how a November 2022 chat product became a global default

Before ChatGPT, OpenAI's models were reachable mainly through an API and a playground. Those interfaces assumed a developer who understood tokens, temperature settings, and prompt structure. ChatGPT wrapped the same underlying model family in a conversational interface with a single text box and a running transcript. That packaging change, not a new model, is what shipped on November 30, 2022.

The training pipeline behind that first release used supervised fine-tuning on human-written demonstrations, followed by reinforcement learning from human feedback. Human raters ranked model outputs, and a reward model learned their preferences. The language model was then optimized against that reward model. This is why the release felt different from raw text completion: the model was tuned to follow instructions and stay in a dialogue, not just continue text.

The interface also imposed server-side limits. The original preview had no browsing, no plugins, and no image input. It could not search the live web, and its knowledge cut off well before launch. OpenAI labeled it a research preview specifically to signal instability and to set expectations that outputs could be wrong.

A useful mental model is that ChatGPT was a product launch built on an existing model, not a model launch. The GPT-3.5 family existed before the chat wrapper. The release date people cite is the date the wrapper became public.

Quantified context: dates, versions, and adoption numbers

The first release is the anchor, but the name "ChatGPT" now covers a family of products with separate release dates. Confusing them is the most common source of wrong answers.

Timeline table:

Date Milestone What changed
Nov 30, 2022 ChatGPT research preview Free web chat on GPT-3.5, no browsing
Early 2023 ChatGPT Plus Paid tier introduced at USD 20 per month
Mar 14, 2023 GPT-4 access in ChatGPT Multimodal input for Plus users, stronger reasoning
May 2023 Mobile apps iOS app, later Android, extending reach beyond desktop
Nov 6, 2023 GPT-4 Turbo and custom GPTs Longer context, lower API pricing, user-built assistants
May 13, 2024 GPT-4o Faster multimodal model, free tier expanded
2024–2025 Reasoning model line Deliberation-style models integrated into ChatGPT tiers

The adoption figure most often quoted is one million users in five days, reported in early December 2022. By January 2023, estimates of monthly active users ranged into the tens of millions. Analysts later described ChatGPT as reaching an estimated 100 million monthly active users within about two months, which would make it one of the fastest consumer applications to that mark. Treat these as third-party estimates, not audited figures.

The original Plus price of USD 20 per month has been remarkably stable. That price point matters because it set the reference for consumer AI subscriptions and anchored competitor pricing for years.

Edge cases: what the release date does not tell you

The date November 30, 2022 is the web launch, not the first internal build. OpenAI had working prototypes earlier. Citing an internal demo date as the release date is a category error.

Regional availability varied. Some countries saw access blocked or delayed, and account creation depended on phone verification and supported regions. A user in one country could have first access weeks or months after the US launch. School and enterprise rollouts lagged further.

The model behind the first preview was not fixed. OpenAI updated the underlying model repeatedly without renaming the product, so two people using "ChatGPT" in early 2023 may have been talking to different snapshots. Version names like GPT-3.5 and GPT-4 refer to model families, while "ChatGPT" refers to the product that hosts them.

Mobile access is a separate date. The iOS app arrived in May 2023, months after the web launch. Anyone whose first exposure was the app experienced a later product than the headline date suggests.

Counterintuitive detail: the release that mattered most was not the most capable model available at the time. GPT-4 arrived months later and scored far higher on standardized exams, yet the November 2022 preview is the event that changed public behavior. Distribution and interface beat raw benchmark scores.

Misconceptions that distort the timeline

The most persistent error is treating ChatGPT as synonymous with GPT-3. GPT-3 was announced in 2020 and was an API product. ChatGPT launched in 2022 on a later, instruction-tuned model. The names are related but the dates are not interchangeable.

A second error is assuming the release date equals the knowledge cutoff. The model could not know events after its training data ended, which was earlier than November 2022. Users who asked about late-2022 news got confident but stale answers.

A third error is believing the free preview was a permanent free tier. OpenAI repeatedly changed what the free tier could access, which model it routed to, and whether logged-out use was allowed. The product named ChatGPT in 2025 is a different service from the one that shipped in 2022, even though the URL and name persist.

Red-flag warnings for anyone citing dates:

Do not cite a model family's announcement date as the ChatGPT product release date. Do not treat third-party user estimates as audited figures; label them as estimates. Do not assume features like browsing, image input, or file upload existed at launch. Do not assume the free tier's capabilities are stable across months. Do not conflate knowledge cutoff with release date; they are different and both shift over time.

Plain-language version: the name stayed the same while almost everything behind it changed.

Practical actions: how to pin down a date you can defend

If you need a defensible date for writing, teaching, or a report, separate the claim into product, model, and feature.

Step one: state the product date. Use November 30, 2022 for the public ChatGPT web release. Attribute it to OpenAI's announcement.

Step two: name the model if it matters. Say "GPT-3.5-based" for the launch build, and give separate dates for GPT-4 access (March 2023) and later models.

Step three: check whether the feature you care about existed at launch. Browsing, plugins, image input, and voice all arrived later. A timeline that lists them as launch features is wrong.

Step four: for adoption claims, use ranges and label the source type. "Reported at roughly one million users in five days" is defensible; a single precise figure presented as fact is not.

Step five: when prices or tier limits matter, verify against current documentation rather than memory. Subscription pricing and free-tier routing change without notice.

Threshold reconciliation: if one source says "launched in 2022" and another says "GPT-4 launched in 2023," both can be correct. They describe different objects. Reconcile by asking whether the claim is about the product, the model, or a feature, then assign the matching date.

Limitations: what the public record cannot settle

OpenAI has not published a complete changelog mapping every model snapshot to every date. Some updates shipped silently. That means any precise claim about which model version served a given user on a given day rests on inference, not documentation.

User counts are estimates assembled by third parties from traffic sampling and surveys. They carry sampling error and definitional differences, such as monthly versus weekly active users. Treat large round numbers as directional.

Finally, the release date tells you when the product became public, not when its capabilities became possible. The underlying research predates launch by years. For historical accuracy, distinguish the announcement, the public availability, and the model training timeline, and cite each separately.