ChatGPT is the consumer AI chatbot from OpenAI and the product that launched the modern generative AI era. Since its launch in November 2022, it has become the fastest-adopted consumer software in history — crossing 100 million weekly active users in early 2023 and reaching more than 800 million by 2026 — and the reference point against which every other AI product is measured.
This page is the living reference to ChatGPT — the product, the underlying models, pricing, features, limitations, and news as it breaks. We update it whenever OpenAI ships a meaningful change, and we link every relevant ChatGPT story we publish here, so it's the one URL you should bookmark if you're trying to stay current without reading every piece we run.
At its core, ChatGPT is a chat interface on top of OpenAI's frontier language model — currently GPT-5.4. You type, it responds. But over the three years since launch, the product has expanded dramatically. It can now see images, speak back in real time, generate images and short video, write and execute code, operate a browser, and remember you across conversations. Most of the interesting work inside ChatGPT today happens via tools the model invokes rather than pure text generation — search, code execution, image generation, browser control, retrieval over connected apps like Google Drive and GitHub.
We rated ChatGPT 4.7 / 5. Still the default AI chatbot for most people, and for good reason — broadest model lineup, deepest integrations, best mobile app. Plus at $20/month is a clear yes; whether you also need Claude or Gemini depends on your specific work.
Read the full ChatGPT review →Latest news about ChatGPT
Every story we've published tagged ChatGPT. Refreshed automatically every time the newsroom publishes.
- Aug 24, 2026Cheshire Academy uses a traffic-light system to govern AI use in classThe 400-student Connecticut boarding school labels every assignment green, yellow, or red — and lets teachers pick the tools.
- Aug 21, 2026ChatGPT gains an Apple Messages plug-in that can draft and send textsOpenAI's new integration lets ChatGPT sort, edit, and send iMessages on a user's behalf — with a local runtime and a warning about auto-approval.
- Aug 21, 2026OpenAI claws back ground on Anthropic among US businesses, Ramp data showsAnthropic still leads at nearly 44% share to OpenAI's 40%, but the ChatGPT maker is now growing faster in Q3 among Ramp's business customers.
- Aug 20, 2026Pew finds 35% of post-ChatGPT web pages show signs of AI authorshipA near-half-million-page Pew analysis says .com sites are 10x more likely to be AI-written than .edu or .gov domains.
- Aug 18, 2026Stanford's AI Observatory finds Anthropic filters out 48% of Claude conversationsA new independent dataset shows sensitive AI use — companionship, health, harassment — runs far higher than company reports admit.
- Aug 18, 2026OpenAI launches ChatGPT for Teens with parental controls and healthy-use limitsA dedicated teen product arrives as OpenAI faces mounting scrutiny over how minors interact with its chatbot.
Timeline of major releases
- Nov 30, 2022ChatGPT launchesOpenAI releases ChatGPT as a free research preview built on GPT-3.5. It crosses one million users in five days and becomes the fastest-growing consumer product in history.
- Feb 1, 2023ChatGPT Plus debutsOpenAI introduces a $20/month subscription offering priority access and faster responses, the template every AI product would eventually copy.
- Mar 14, 2023GPT-4 releasedA substantial jump in reasoning and multimodal capability, initially available only to Plus subscribers via the ChatGPT interface.
- Nov 6, 2023Custom GPTs and the GPT StoreUsers can build their own ChatGPTs from a prompt plus tools. The store later opens for revenue sharing with creators.
- May 13, 2024GPT-4o ("omni") launchesFirst real-time voice and vision model with dramatically lower latency. Free tier gets access to a scaled-down version, eroding Claude and Gemini's free-tier advantage.
- Sep 12, 2024o1 reasoning models previewOpenAI's first explicit reasoning model — trades latency for correctness on hard problems. Starts the "thinking models" category.
- Aug 6, 2025GPT-5 shipsUnified frontier model that routes between fast and reasoning modes automatically. Default model for Plus and Free tiers.
- Apr 24, 2026GPT-5.4 and Agent mode general availabilityLonger autonomous task horizons, sharper coding, and Agent mode (browser + code execution) goes GA for Plus, Pro, and Team.
How to access ChatGPT
| Plan | Price | Includes |
|---|---|---|
| Free | $0 | GPT-5 with reduced message cap, limited voice and image input, standard response speed. |
| Plus | $20/mo | Higher GPT-5 caps, GPT-5 Thinking, Agent mode, Advanced Voice, image generation, 32K context. |
| Pro | $200/mo | Unlimited GPT-5 Thinking, priority on new models, research-grade Deep Research, 128K context. |
| Team | $30/user/mo | Plus features plus shared workspace, admin console, no-training data promise. |
| Enterprise | Contact sales | Custom deployment, SSO/SCIM, unlimited high-speed GPT-5, SOC 2 Type II, dedicated TPM. |
ChatGPT vs. Claude vs. Gemini
| Feature | ChatGPT | Claude | Gemini |
|---|---|---|---|
| Free tier available | Yes (capped GPT-5) | Yes (capped Claude Sonnet) | Yes (Gemini 2.5 Flash) |
| Consumer price (entry paid tier) | $20/mo Plus | $20/mo Pro | $20/mo AI Premium (via Google One) |
| Context window (paid) | 128K | 200K | 1M+ (2M in Advanced) |
| Native image generation | Yes (DALL-E 3 / GPT Image) | No (text output only) | Yes (Imagen) |
| Real-time voice | Advanced Voice (GPT-5o) | No | Gemini Live |
| Agent / autonomous mode | Agent mode (GA) | Computer Use (beta) | Project Astra (limited) |
| Code execution | Yes (sandbox) | Yes (via Artifacts / Claude Code) | Yes (Canvas) |
| API access | Yes (openai.com/api) | Yes (console.anthropic.com) | Yes (aistudio.google.com) |
How ChatGPT works under the hood
ChatGPT is what OpenAI calls a "thin product layer" on top of its models. When you send a message, a routing system decides which underlying model to call: GPT-5 for most queries, GPT-5 Thinking for harder reasoning, a faster small model for trivial questions, an image model for pictures, a voice model for speech. The right answer for a given prompt might involve two or three of these in sequence — you don't see that, you just see the reply.
This "router" approach is why ChatGPT often feels faster than calling a single model via the API. OpenAI is betting that most users don't want to choose models — they want the product to choose for them. The tradeoff is less transparency: two identical prompts can take different paths and get different-quality answers, which is something we flag when reviewing ChatGPT's consistency for specific tasks.
The underlying GPT-5 model family was released in August 2025 and is the default for nearly every ChatGPT user. Unlike prior generations, GPT-5 unifies what used to be separate "fast" and "reasoning" models into one system that decides on the fly how much thinking to apply. For power users, Plus and Pro expose GPT-5 Thinking as a separate, slower, more thorough mode; for Free users, the router controls everything.
What ChatGPT is actually good at
Most of our readers have used ChatGPT for drafting, summarizing, and quick questions. The less obvious wins are coding assistance, research synthesis, and repetitive knowledge work — any task where the bottleneck is producing a first draft of something structured. Since the mid-2024 release of GPT-4o and then the 2025 move to GPT-5, ChatGPT has been genuinely useful for writing production-quality code if you supervise it, and for turning messy input (transcripts, emails, notes) into clean structured output.
It's less good at things it can't verify. It will confidently state incorrect facts about obscure topics, make up URLs, invent citations, and smooth over real-world complexity that doesn't fit the template of its training data. Agent mode mitigates some of this — when the model can browse and cite sources, hallucinations drop substantially — but for anything mission-critical, the pattern we recommend is "AI drafts, human verifies," not "AI decides."
Our AI tools comparison guide goes deeper on the head-to-head against Claude and Gemini for specific tasks.
Pricing and plans
OpenAI has five tiers. Free is good for casual use, Plus ($20/month) is the right default for anyone using ChatGPT regularly for work, Pro ($200/month) is aimed at power users and small businesses, and Team/Enterprise are for organizations that need admin controls and data guarantees. See the pricing table above for what's in each tier.
The pricing has been stable at $20/month for Plus since February 2023, which is notable in an industry where most competitors have repeatedly raised prices. Pro, which launched at $200/month in December 2024, has also held steady. OpenAI's bet is that it can hold prices while increasing capability, building loyalty the way consumer brands do, rather than squeezing power users for every marginal dollar of willingness to pay.
Privacy, data, and training
The default for Free and Plus chats is that OpenAI may use them to improve future models. You can turn this off in Data Controls — doing so disables chat history retention for those chats but doesn't affect model quality for you. Team, Enterprise, and API traffic is not used for training by default, and Enterprise customers can negotiate zero-retention agreements where prompts and responses aren't stored at all after the session.
OpenAI publishes a usage policy that prohibits certain uses (weapons, targeted harassment, electoral interference, and so on) and enforces it with a mix of automated detection and account review. The line for what's acceptable in ChatGPT has moved over time, generally toward more permissive defaults for adult users, with some tightening around real-person impersonation.
ChatGPT for business and developers
OpenAI's business traffic now comes from two surfaces: ChatGPT Team/Enterprise for employees, and the API for developers. The API is where you'd integrate ChatGPT's underlying models into your own software — Zapier, Notion, and thousands of smaller products are built on it. It's priced by token (roughly, by word) and scales from pennies per interaction at the low end to dollars at the high end for the biggest reasoning models.
If you're evaluating ChatGPT for a business, our AI agents guide is the deeper read on Agent mode and what autonomous workflows actually look like in production. The short version: Agent mode works well for well-defined, bounded tasks — research, data pulling, form filling, simple browser automation — and gets flaky on anything requiring judgment about trust or ambiguity. A human-in-the-loop pattern still outperforms full autonomy for most business-critical work.
Controversies and limitations
ChatGPT has been at the center of several recurring controversies since launch: the use of copyrighted text in its training data (the subject of major lawsuits from the NYT and others), hallucinations in high-stakes settings (a lawyer was sanctioned for citing fake cases from ChatGPT), accessibility in regulated industries (healthcare, legal, government), and the political lean of its default responses. We cover these as they evolve on our ChatGPT news tag.
The technical limitations worth knowing: it has a training cutoff (it doesn't know what happened after a certain date unless it browses), it can't reliably do precise math without the code execution tool, it can be jailbroken with enough effort, and it performs worse in non-English languages — especially low-resource ones — even though it's much better than a year ago.
Our coverage
Everything we write tagged ChatGPT appears above in Latest news. For the underlying company, see OpenAI. For the biggest alternatives, see Claude and Gemini. For how ChatGPT fits alongside other AI products, see our AI tools comparison guide.
