March 2024, a tax consultant's office in Buenos Aires. Until that month, every time a client messaged "hey, how does an LLC that sells services abroad invoice?", the consultant had to open the tax authority's PDF, find the article, check the latest resolution. Fifteen minutes per question, several a day.
That same week he signed up for ChatGPT Plus, loaded a custom GPT with the full text of the national small-business tax regime, the most recent general resolutions, and the City of Buenos Aires tax code. He wrote instructions for it: "always cite the article and the norm, and if you're not sure say so." Today that GPT is the first-pass filter for every client question. The human answer comes after, but it starts with 80 percent of the work already done.
That story — repeated tens of thousands of times through 2024 and 2025 — explains better than any pitch why ChatGPT stopped being a product and became a platform.
How we got here
Quick recap. ChatGPT launched in November 2022 as a text box with a language model behind it. Two months later it hit a hundred million users — a record for consumer adoption. We cover that history in detail in the OpenAI piece.
What happened next is that OpenAI kept stacking layers on top of that text box. First DALL-E for images, then Voice mode, then Canvas, then custom GPTs, then Advanced Voice, then Operator and Sora. In every release, the internal question wasn't "how do we improve the model?" but "what else can the platform do?"
As of April 2026, ChatGPT has around 300 million weekly active users (figure confirmed by OpenAI in December 2024 and reported by Reuters). It's by far the most used AI application in the world.
What it offers today
Canvas. OpenAI's answer to Claude's Artifacts. A side panel where the document or code you're working on sits apart from the chat. You edit directly, ChatGPT suggests targeted changes, and the chat becomes the instruction space. For long-form writing and code review, it's noticeably more comfortable than a linear chat.
Native web search. ChatGPT searches the internet in real time. You ask it for today's exchange rate, for someone's latest press conference, for the week's news, and it answers with cited sources. Claude now has search too, but ChatGPT shipped earlier and the experience is more polished.
Persistent memory. Since 2024, ChatGPT remembers things across conversations — your profession, your tone preferences, projects you're working on. You can turn it on, off, and edit what it remembers. Handy for continuous use.
Advanced Data Analysis (previously Code Interpreter). You upload a CSV, an Excel file, a PDF with tables, and you say "analyze this, pull the key metrics and make me the charts." ChatGPT runs Python in a sandbox and gives you results with visualizations. No coding required.
GPT Store and your own GPTs. The GPT Store has thousands of specialized assistants built by third parties. You go in, search "LinkedIn copywriter," find twenty, try the top-ranked. And if there isn't one for your thing, you build your own: you give it instructions in plain language, upload knowledge files, and optionally configure actions (calls to external APIs). No code. In fifteen minutes you have your own assistant.
Advanced Voice Mode. Launched September 2024. Real-time spoken conversation — you interrupt, change subject, it laughs, shifts tone. It's honestly the most natural voice interface on the consumer market. Useful for brainstorming while walking, practicing a language, thinking out loud.
ChatGPT in WhatsApp and Apple Intelligence. You can message the official ChatGPT number on WhatsApp and chat without leaving the app. And since iOS 18, Siri can delegate complex questions to ChatGPT directly — integration done with OpenAI as the primary partner.
Where it's worth it and where it isn't
Here's an honest read.
ChatGPT wins, by a margin, on image generation. DALL-E 3 inside the chat, with conversational iteration, is the most comfortable path that exists to an image that actually works. Claude doesn't generate images. Gemini generates but with less controllable results.
It also wins on voice. Advanced Voice Mode is ahead of what Google and Anthropic have today for real-time spoken conversation. It's a visible difference, not a marketing one.
It wins on ecosystem breadth. GPT Store, distribution inside Microsoft 365, WhatsApp and Apple integration, presence in hundreds of products via API. If you're looking for "an AI that's already where you're working," ChatGPT is the most likely to deliver.
Claude wins, also by a margin, on coherent long-form writing, on large-PDF analysis, on literal following of complex instructions, on production code, and on predictability of refusals. For professional work with sensitive data, Claude is the more defensible choice.
Gemini wins on native Google product integration — if you live in Gmail, Docs, and Drive, having Gemini built in there is a structural advantage.
Plus, Team, Enterprise — which one fits
ChatGPT Plus costs $20 a month. It opens up Advanced Voice without limits, DALL-E 3 without quota, the o3/o4 reasoning models, Canvas, memory, Deep Research, and the ability to build your own GPTs. For a professional using it daily, it pays for itself.
ChatGPT Team is for small teams: $25-$30 per user per month with shared workspace, private team GPTs, and no training on your data.
ChatGPT Enterprise is for large companies: custom pricing, full admin controls, SSO, and privacy and audit guarantees. This line grew fast through 2024-2025, with customers including PwC, Moderna, Klarna, and several governments.
To close, and to keep going
The question worth asking isn't "which AI is best?" but "which one is my default, and which ones are my specialists?"
For many professionals working on writing, code, and sensitive analysis, that default is Claude and ChatGPT becomes the specialist for images, voice, and creative exploration. For many other people — especially those doing broader work and less critical production — the default is ChatGPT and it's worth having as the main tool.
Neither read is wrong. What doesn't work is picking one and stopping the exploration: the field moves fast and the trade-offs shift every few months.
If you want to place ChatGPT in the wider competitive map, The AI race gives you the picture. If you want the full history of how OpenAI got here, OpenAI and ChatGPT — the one that turned the lights on is the companion read.
What role does ChatGPT play in your stack today: primary default, or specialist for specific tasks?
A twelve-year-old sitting at the kitchen table with his phone propped against the pitcher. He types into ChatGPT: "draw me a picture of Rocco, my dog, but as a superhero flying over a city at night." Fifteen seconds later, the image shows up on screen: a beagle with a red cape, yellow eyes, skyscraper lights behind him. The kid laughs. Sends it to his mom on WhatsApp.
That scene, repeated millions of times a day, is a good way into what ChatGPT is today.
What it is, what it does
ChatGPT is the most used AI application in the world. You open chat.openai.com or the app and you get a text box. You type whatever — a question, a task, a creative request — and it answers.
What made it unique is that it was the first AI anyone could use without knowing any tech. We covered that history in another piece. What matters here is what it does now.
Five things worth knowing
Basic chat. You ask, it answers. You can write in any language and it replies in that language. Works for translating, summarizing, explaining, fixing a draft, writing an email, planning a trip.
You can upload photos, files, and audio. You drag a picture into the chat box and it describes or analyzes it. You upload a PDF and it summarizes it. You talk through the mic and it understands.
It generates images with DALL-E 3. The dog-superhero thing above? That's a real image DALL-E 3 made inside ChatGPT. You describe what you want in plain language and it draws it. Useful for fun, and also for covers, illustrations, quick mockups.
Basic voice mode. You press a button and talk to it like a phone call. It listens, answers in a natural voice, and you can hold a conversation. Handy while driving or cooking.
GPTs from the store. In the menu there's a section called "Explore GPTs." These are specialized assistants other people or companies built: one that helps you write sales emails, another that's a vegetarian-recipe expert, another that puts together workout routines. You pick them and use them like mini-apps inside ChatGPT.
Free vs Plus — what to pay for and what not to
ChatGPT has two main tiers for individual users.
Free (no cost): gives you access to most things with daily quotas. If you use it a bit each day, it's enough. You can ask questions, upload photos, generate some images.
Plus ($20 a month): removes the limits. If you use it several times a day, if you want to generate images without restrictions, if you want the good advanced voice mode, paying for it is worth it fast.
No need to start on Plus. Try Free for a week and see whether you hit the limits.
What to take away
Three practical things to get you started:
- Write to it the way you talk to a person. ChatGPT doesn't need special commands or weird formatting. You explain what you want in your own words. If something didn't come out right, you tell it "no, redo that but shorter" or "change this."
- Try DALL-E 3 at least once. It's the most fun way to realize what a modern AI can do. Ask it for a weird, impossible picture of something personal. It'll make it.
- For serious tasks, look at other tools too. ChatGPT is the most complete one to start with and for general use. For high-quality long-form writing or for production code, Claude is often the more predictable pick. There's no single "best AI" — there are better ones for each task.
March 2024, a tax consultant's office in Buenos Aires. Until that month, every time a client messaged "hey, how does an LLC that sells services abroad invoice?", the consultant had to open the tax authority's PDF, find the article, check the latest resolution. Fifteen minutes per question, several a day.
That same week he signed up for ChatGPT Plus, loaded a custom GPT with the full text of the national small-business tax regime, the most recent general resolutions, and the City of Buenos Aires tax code. He wrote instructions for it: "always cite the article and the norm, and if you're not sure say so." Today that GPT is the first-pass filter for every client question. The human answer comes after, but it starts with 80 percent of the work already done.
That story — repeated tens of thousands of times through 2024 and 2025 — explains better than any pitch why ChatGPT stopped being a product and became a platform.
How we got here
Quick recap. ChatGPT launched in November 2022 as a text box with a language model behind it. Two months later it hit a hundred million users — a record for consumer adoption. We cover that history in detail in the OpenAI piece.
What happened next is that OpenAI kept stacking layers on top of that text box. First DALL-E for images, then Voice mode, then Canvas, then custom GPTs, then Advanced Voice, then Operator and Sora. In every release, the internal question wasn't "how do we improve the model?" but "what else can the platform do?"
As of April 2026, ChatGPT has around 300 million weekly active users (figure confirmed by OpenAI in December 2024 and reported by Reuters). It's by far the most used AI application in the world.
What it offers today
Canvas. OpenAI's answer to Claude's Artifacts. A side panel where the document or code you're working on sits apart from the chat. You edit directly, ChatGPT suggests targeted changes, and the chat becomes the instruction space. For long-form writing and code review, it's noticeably more comfortable than a linear chat.
Native web search. ChatGPT searches the internet in real time. You ask it for today's exchange rate, for someone's latest press conference, for the week's news, and it answers with cited sources. Claude now has search too, but ChatGPT shipped earlier and the experience is more polished.
Persistent memory. Since 2024, ChatGPT remembers things across conversations — your profession, your tone preferences, projects you're working on. You can turn it on, off, and edit what it remembers. Handy for continuous use.
Advanced Data Analysis (previously Code Interpreter). You upload a CSV, an Excel file, a PDF with tables, and you say "analyze this, pull the key metrics and make me the charts." ChatGPT runs Python in a sandbox and gives you results with visualizations. No coding required.
GPT Store and your own GPTs. The GPT Store has thousands of specialized assistants built by third parties. You go in, search "LinkedIn copywriter," find twenty, try the top-ranked. And if there isn't one for your thing, you build your own: you give it instructions in plain language, upload knowledge files, and optionally configure actions (calls to external APIs). No code. In fifteen minutes you have your own assistant.
Advanced Voice Mode. Launched September 2024. Real-time spoken conversation — you interrupt, change subject, it laughs, shifts tone. It's honestly the most natural voice interface on the consumer market. Useful for brainstorming while walking, practicing a language, thinking out loud.
ChatGPT in WhatsApp and Apple Intelligence. You can message the official ChatGPT number on WhatsApp and chat without leaving the app. And since iOS 18, Siri can delegate complex questions to ChatGPT directly — integration done with OpenAI as the primary partner.
Where it's worth it and where it isn't
Here's an honest read.
ChatGPT wins, by a margin, on image generation. DALL-E 3 inside the chat, with conversational iteration, is the most comfortable path that exists to an image that actually works. Claude doesn't generate images. Gemini generates but with less controllable results.
It also wins on voice. Advanced Voice Mode is ahead of what Google and Anthropic have today for real-time spoken conversation. It's a visible difference, not a marketing one.
It wins on ecosystem breadth. GPT Store, distribution inside Microsoft 365, WhatsApp and Apple integration, presence in hundreds of products via API. If you're looking for "an AI that's already where you're working," ChatGPT is the most likely to deliver.
Claude wins, also by a margin, on coherent long-form writing, on large-PDF analysis, on literal following of complex instructions, on production code, and on predictability of refusals. For professional work with sensitive data, Claude is the more defensible choice.
Gemini wins on native Google product integration — if you live in Gmail, Docs, and Drive, having Gemini built in there is a structural advantage.
Plus, Team, Enterprise — which one fits
ChatGPT Plus costs $20 a month. It opens up Advanced Voice without limits, DALL-E 3 without quota, the o3/o4 reasoning models, Canvas, memory, Deep Research, and the ability to build your own GPTs. For a professional using it daily, it pays for itself.
ChatGPT Team is for small teams: $25-$30 per user per month with shared workspace, private team GPTs, and no training on your data.
ChatGPT Enterprise is for large companies: custom pricing, full admin controls, SSO, and privacy and audit guarantees. This line grew fast through 2024-2025, with customers including PwC, Moderna, Klarna, and several governments.
To close, and to keep going
The question worth asking isn't "which AI is best?" but "which one is my default, and which ones are my specialists?"
For many professionals working on writing, code, and sensitive analysis, that default is Claude and ChatGPT becomes the specialist for images, voice, and creative exploration. For many other people — especially those doing broader work and less critical production — the default is ChatGPT and it's worth having as the main tool.
Neither read is wrong. What doesn't work is picking one and stopping the exploration: the field moves fast and the trade-offs shift every few months.
If you want to place ChatGPT in the wider competitive map, The AI race gives you the picture. If you want the full history of how OpenAI got here, OpenAI and ChatGPT — the one that turned the lights on is the companion read.
What role does ChatGPT play in your stack today: primary default, or specialist for specific tasks?
On September 24, 2024, during DevDay in San Francisco, Mira Murati — OpenAI's CTO at the time — fired up Advanced Voice Mode live in front of the audience, interrupted the model three times in the first twenty seconds, asked it to tell a story in a dramatic tone, then in a whispered tone, then in a sports-announcer tone. The model switched all three times without pause, with natural prosody, with laughs inserted where they fit. Google and Anthropic didn't have at that moment — and wouldn't have in the following months — a comparable consumer-production implementation.
That moment holds an uncomfortable operational truth for those of us who prefer Claude as a professional production tool: on the layer of breadth-of-experience for the end user, OpenAI is ahead. And that distance didn't close in the eighteen months that followed. Worth taking apart why.
What's underneath the stack
To understand the surface you have to see the technical structure. ChatGPT in April 2026 isn't a model, it's a set of models orchestrated behind a common interface.
GPT-4o is the default multimodal model — text, image, and audio in a single request. Low latency, low cost, most non-paying users use it without knowing.
o3 and o4 are the reasoning family. They spend extra compute on an internal chain of thought before producing the final answer. On Olympic-math and competitive-programming benchmarks, o3 approaches elite human level. The cost is latency: a hard problem can take thirty seconds.
DALL-E 3 is the image generator. Built natively into chat — you describe, iterate, refine in plain language. The operational advantage over Midjourney isn't raw quality (Midjourney often has better aesthetic control) but minimum friction.
Sora is the video generator. Released to the general public in December 2024. It produces clips up to a minute long with reasonable temporal and spatial coherence. It's not the best pure video model — Runway and Google Veo compete closely — but it's integrated into the same ecosystem as everything else.
Realtime API lets developers build their own applications that consume the voice pipeline in real time without going through the ChatGPT interface. It's what enables third parties to build their own voice experiences on top of OpenAI infrastructure.
Agents SDK (the evolution of the Assistants API) is the agent-building framework: state, tools, memory, orchestration across multiple models.
The architectural thesis behind the stack is singular: don't compete on better model, compete on better platform. OpenAI bet that the value would get captured by whoever had the sum of integrated capabilities, not by whoever had the marginally better text model. So far, that bet is paying off.
Operator and Deep Research — the frontier products
Two products launched between January and February 2025 define OpenAI's current product-level competitive frontier.
Operator (January 2025) is an agent that browses the web in a dedicated browser, executes complex task flows — booking flights, making purchases, filling forms, pulling data across multiple sites — and returns results. What sets Operator apart from earlier alternatives is the combination of vision (it processes browser screenshots), reasoning (it decides the next step), and execution (it emits commands). Claude has had Computer Use since October 2024 with a conceptually similar architecture; the difference is that Operator is packaged as a consumer product with polished UX, while Computer Use remains primarily an API capability for developers.
Deep Research (February 2025) is a research mode that takes a complex question ("write me a 20-page report on the sodium battery market in Asia"), navigates autonomously through tens or hundreds of sources over five to thirty minutes, and returns a cited report. It's built on top of the o-series reasoning models and is the most concrete demonstration to date of what a well-integrated research agent can do.
Both products require Plus at minimum, and Deep Research with expanded quota requires the Pro plan ($200 a month).
Where OpenAI wins: thesis by layer
Worth a technical breakdown of exactly where OpenAI is superior, because the generality of "ChatGPT is bigger" doesn't support professional decisions.
Visual generation layer. DALL-E 3 integrated + Sora + the multimodal quality of GPT-4o processing images as input. Claude doesn't generate images or video. Gemini generates but with less consistent results for professional use. For any flow that requires moving quickly from idea to image or video, ChatGPT is the lowest-friction option.
Voice layer. Advanced Voice Mode + Realtime API. It's territory where OpenAI has a generation of lead. Google announced Gemini Live in May 2024 but the current implementation isn't at the same level. Anthropic has no native voice offering in consumer production.
Distribution layer. GPT Store + Microsoft 365 Copilot + WhatsApp + Apple Intelligence + hundreds of API integrations. The advantage isn't technical; it's structural. OpenAI signed as Apple's primary partner for Siri delegation, secured native WhatsApp integration, and maintains the privileged Microsoft relationship that gives it access to the largest corporate base in the world.
Specialized agent ecosystem breadth layer. The GPT Store reached several million third-party-built GPTs by 2026. Most aren't good — quality distribution is extreme Pareto. But the absolute mass creates network effects competitors find hard to replicate.
Where OpenAI doesn't win: honest thesis
And the inverse breakdown is worth doing too.
Long-form quality writing layer. Claude is consistently superior in extended coherent writing, in holding tone and argument across thousands of words, in producing text that doesn't fall apart on careful rereading. There's no single academic benchmark that captures this — it's a property that becomes obvious when you run both models on the same long-form prompt.
Large-document analysis layer. Claude Opus 4.7 handles extended context windows with coherence that ChatGPT today doesn't match in real work on hundred-page PDFs. Anthropic's "long context" engineering remains state of the art.
Production code layer. Claude Code, combined with the base model in Opus 4.7, is the tool most serious developers pick for sustained work in large repos. ChatGPT with GPT-4o or o3 is competitive for discrete tasks but not for continuous production work.
Refusal predictability layer. Claude refuses consistently and explainably; ChatGPT has a documented history of erratic refusals that change between versions. For professional work on sensitive topics, that unpredictability is an operating cost.
Native Google Workspace integration layer. If your company lives in Gmail, Docs, Drive, Calendar — Gemini has a structural advantage. Not on technical merit but on privileged surface access.
Pricing and segmentation
OpenAI's price line as of April 2026 is this.
Free. GPT-4o with quotas, basic web search, limited images. Surprisingly capable for casual use.
Plus ($20/month). Everything on Free without hard quotas + o3/o4, Canvas, Advanced Voice, unlimited DALL-E, GPT building, Deep Research with quota, persistent memory.
Pro ($200/month). Plus + Deep Research with expanded quota + o1-pro (a reinforced version of the reasoning model) + Operator without restrictions + Sora with higher quotas. A tier clearly targeted at research professionals and heavy developers.
Team ($25-$30/user/month). Shared workspace, private team GPTs, no training on your data.
Enterprise (custom pricing). Admin controls, SSO, audit guarantees, dedicated support. Grew to thousands of corporate customers between 2024 and 2026.
The pricing strategy reflects the product thesis: capture the general user with Plus, capture the professional with Pro, capture the corporation with Enterprise. Each tier unlocks differentiated tools — it isn't just "more quota of the same thing."
My editorial thesis
I'll close with a read I allow myself because of who I am — a consultant who uses Claude as his primary production tool and who teaches Claude in a course that carries his last name.
If you have to recommend a single AI tool in 2026 to a regular person, without technical context or critical-work needs, ChatGPT remains the most defensible answer. On capability breadth, on distribution, on how much it resolves in one place, on ease of entry, on a GPT ecosystem that covers use cases you wouldn't have thought of. That's the Swiss Army Knife of AI — it has the scissors, the screwdriver, the magnifier, the corkscrew, and it all fits in your pocket.
If the person in question is a professional and is going to use AI as a production tool on work where quality and predictability of output matter — long-form writing, legal or financial analysis on large documents, production code, decisions on sensitive data — Claude is the more defensible pick. Not because it's worse on breadth, but because it's better at the specific task that defines that work. That's the scalpel — it does one thing very well, very consistently, and when precision matters there's no substitute.
The choice between Swiss Army Knife and scalpel isn't a choice about which is "better." It's a choice about which surgery you're doing. What doesn't make sense is ignoring that both tools exist and picking by tribal loyalty to one of the two brands. The professional doing good work in 2026 probably pays both twenty-dollar subscriptions a month and knows exactly when to open each app.
What's the empirical test you use to decide, faced with a new task, whether you open ChatGPT or Claude first?
