ChatGPT

Where ChatGPT sits

ChatGPT is the product that made this entire category mainstream. OpenAI released it on 30 November 2022 as a "low-key research preview," and by February 2026 OpenAI was reporting more than 900 million weekly active users and over 50 million paying consumer subscribers, alongside roughly 9 million business subscribers. Whatever else is true in this market, ChatGPT is the default — the tool non-technical friends mean when they say "AI," the one with the largest integration footprint, and the one every competitor is priced against. That scale is both the product's asset and its burden: features arrive here first, models change here fastest, and the plan structure has grown complicated enough that choosing wrong costs real money.

The 2026 lineup is six tiers. Free at $0 with unlimited text chats on the entry model and caps on uploads, images, voice, deep research and Codex. Go at $8 a month for more of everything but no advanced reasoning, and — the caveat OpenAI itself prints — ads may appear on it (they have also been tested on Free in the US). Plus at $20 a month, monthly only, which is the tier that brings in reasoning models, projects, scheduled tasks, custom GPTs, Sites, and expanded Codex and ChatGPT Work. Pro comes in three usage tiers — $100, $200 and $500 a month — with the top tier adding an "Ultrafast" speed setting and new Pro 200 sign-ups receiving a smaller usage allowance than earlier subscribers kept (grandfathered through 29 October 2026). Business seats are $25 per user monthly ($20 billed annually, two-seat minimum) for Standard and $125 ($100 annually) for Premium with 5x usage and no 5-hour cap; Enterprise is quote-only with data residency in ten regions, SCIM and custom retention.

What it actually does

The breadth is the point. ChatGPT writes and edits, reasons through problems with configurable thinking levels, searches the web, analyses files and images, generates and edits images, speaks and listens in real-time voice, schedules recurring tasks, runs deep research reports with citations, builds custom GPTs and shared workspace agents, generates and edits video through Sora-integrated experiences where available, and drives an agentic coding product (Codex) and a desktop work surface (ChatGPT Work). Memory across conversations plus "Connectors" to Google Drive, Slack, GitHub and similar give it company context on Business and above. The model roster moves quickly — OpenAI's 5.x and 6 generation models, with names and reasoning tiers varying by plan — and the plan you hold determines which reasoning level you can select.

What sets it apart from Claude, its closest rival, is multimodality and reach: image generation and editing are genuinely good, voice is a real interface rather than a checkbox, and the GPT ecosystem means someone has usually built the odd-shaped tool you need. What Claude does better, in my use, is long-document care and editing restraint; what ChatGPT does better is everything visual, vocal and ecosystem-shaped.

My testing notes

I have run ChatGPT daily since launch across every consumer tier, and periodically on Business workspaces.

First observation: the free tier is a genuine product, and the ceiling is exactly where you would put it if you were monetising. Text chat on the entry model handles ordinary questions, drafts and summaries well enough that occasional users need nothing else. In my testing the walls appear on the second-order features — a third upload in an hour gets a wait, image generation slows and dims in quality, deep research runs rationed — and when a cap lands, the downgrade is quiet rather than announced. For evaluating the product the free tier is honest; for daily professional work it is a demo of Plus. The $8 Go tier sat awkwardly for me: more capacity, but no advanced reasoning and possible ads, so the $12 jump to Plus buys the features that actually change work.

Second observation: Plus at $20 is still the value anchor of this industry, and it has held that price for years while the models behind it changed repeatedly. My standard mixed workload — a research question with citations, a 2,000-word draft with three revision passes, an image for a slide, a scheduled daily digest — all ran without hitting visible caps in a normal working week. Deep research is the feature I use most: a recent run produced a 12-source competitor scan in about eight minutes that I had to correct in two places (both were citation attributions that overreached the source), but the structure and coverage were worth editing rather than redoing. Scheduled Tasks ran reliably; memory was genuinely useful and occasionally surprising in what it chose to retain.

Third observation: on a Business workspace, the difference is governance, not intelligence. Same models, but no training on workspace data by default, SSO, an admin console, and workspace-wide shared GPTs and knowledge. The two-seat minimum is trivial friction; the seat-type mixing (Standard for most, Premium for the one person running long agent sessions) is a sensible pattern I recommend. On reliability: across years of daily use the service is generally dependable, with occasional high-traffic degradations; image generation quality has fluctuated notably across model generations — the current editing tools are much stronger at text rendering inside images than anything I tested two years ago.

What it does not do

ChatGPT will not be the most careful editor of your long document — its tone drifts toward confident verbosity and its citations need checking — and it is not private by default on the consumer tiers: Free, Go, Plus and Pro conversations can be used for model training unless you opt out, and none of those tiers carry SOC 2 or SSO. The plan ladder has traps: the $100-to-$200 jump buys capacity, not capability; new Pro 200 subscribers got a reduced allowance at the same price; and Pro is monthly-only with no annual discount anywhere in the consumer lineup. Ads have entered the cheapest tiers in the US, which changes the data relationship on a paid product. Usage caps are enforced by silent downgrade to weaker models, which can produce a sudden quality drop you will only notice by result. Content policy refusals still appear on legitimate work (medical, legal, security-adjacent), and the very pace of model and feature change means docs and blog posts about ChatGPT are stale within months — including, possibly, parts of this review by the time you read it.

Verdict

ChatGPT is the most complete AI assistant you can buy: nothing matches its combination of text, reasoning, images, voice, agents, scheduling and ecosystem, and Plus at $20 remains the benchmark every competitor is priced against. I rate it 4.5 on the rubric — the deepest core functionality in the category, strong everyday value on Plus, the widest third-party ecosystem and enormous reliability investment — held back from a perfect score by privacy defaults that require opting out, silent limit downgrades, ads on the cheap tiers, a Pro ladder whose capacity-only tiers are easy to overspend on, and an editing temperament that needs supervision on careful work. For most people this is the first AI subscription to buy and often the only one; pair it with a second tool when the work demands documents treated with care.

Sources

• https://chatgpt.com/pricing
• https://openai.com/index/chatgpt/
• https://openai.com/index/testing-ads-in-chatgpt/
• https://openai.com/signals/research/2026q1-update/
• https://techcrunch.com/2026/02/27/chatgpt-reaches-900m-weekly-active-users/
• https://finance.yahoo.com/news/chatgpt-almost-1-billion-weekly-212157499.html?fr=sycsrp_catchall
• https://magai.co/chatgpt-pricing
• https://skillmammoth.com/blog/chatgpt-pricing
• https://fritz.ai/chatgpt-free-vs-paid
• https://nocode.mba/articles/chatgpt-pricing
• https://www.lindy.ai/blog/chatgpt-pricing
• https://en.wikipedia.org/wiki/OpenAI