Three subscriptions. Twenty dollars each. That is the strange position anyone choosing an AI assistant now finds themselves in: ChatGPT Plus, Claude Pro and Google's AI Pro plan all sit at the same monthly price in the United States, so the money cannot make the decision for you. What separates them is not the sticker. It is what each company chose to build around the chat box — and which of those extras matches the work you actually do.

Start with the prices, because they anchor everything else. OpenAI's help centre lists ChatGPT Plus at $20 a month, billed monthly, with no annual option. Anthropic lists Claude Pro at $20 a month, or $17 a month if you pay $200 for a year up front. Google lists its AI Pro plan at $19.99 a month in the US, sold through Google One. Prices vary by country and taxes may be added, but at list price the three are, for practical purposes, identical. So compare the boxes, not the bills.

ChatGPT Plus: the broadest toolbox

Plus is the generalist's pick, and OpenAI's own description reads like a checklist of everything a consumer AI product has accumulated. For the $20 you get priority access when demand is high, higher message limits and broader model choice than the free tier, faster responses, voice conversations, image generation, file uploads with analysis, and Deep Research where it is available. Custom GPTs — small, purpose-built versions of the assistant that users create and share — remain a Plus-centred feature, and the plan also reaches into connected tools through Sign in with ChatGPT.

Two cautions belong here. Model names inside ChatGPT change often enough that OpenAI itself tells subscribers to check the model picker on their own account rather than trust any published list. And API use is a separate bill; a Plus subscription buys you nothing in the API, a point that surprises people who assumed the two were one wallet. If your day mixes writing, spreadsheets, images, voice notes and the odd bit of code, Plus covers the widest ground in a single payment. Readers comparing how the underlying models got here may enjoy our earlier look at the race between the big reasoning models.

Claude Pro: built around writing and long work

Claude's following grew out of a reputation for prose. Long documents go in, careful summaries and drafts come out, and the tone needs less sanding down than most rivals. Anthropic's Pro tier is priced to match Plus almost exactly — $20 monthly, $17 a month on the annual plan — and Anthropic states that Pro gives at least five times more usage per five-hour session than its free plan, with limits resetting on a rolling window and weekly caps on top.

What Pro adds tells you who it is aimed at. Projects let you organise chats and documents around a piece of work and keep them together. Claude Code, Anthropic's terminal-based coding tool, is included in every paid plan and draws on the same usage pool as your chats, which heavy coders should note before a long session. Pro also opens access to more of Anthropic's models, plus Claude in Chrome and Microsoft 365. There is no built-in image generation story comparable to ChatGPT's, and that absence is the honest trade: Pro is a desk for text and code, not a studio. If most of your AI time is spent drafting, editing, reviewing contracts or working through long files, this is the subscription shaped like your day.

The limits deserve a straight read. Five times the free usage sounds generous until a deadline week eats it, and the weekly caps sit on top of the five-hour windows whether you notice them or not. Anthropic is the most explicit of the three about how its limits work, which is a virtue — nobody likes discovering a cap mid-task. Writers who live inside long documents will feel the fit immediately. Anyone who also needs images made inside the same subscription will feel the gap just as fast.

Google AI Pro: the ecosystem bundle

Google sells something different in kind. AI Pro is a Google One plan, so the $19.99 buys Gemini and a bundle wrapped around it: 5 TB of cloud storage across Gmail, Drive and Photos, family sharing with up to five other people, and Gemini placed inside the Google apps themselves — help with proofreading in Gmail, assistance in Docs and Sheets — rather than only in a separate chat window. Google lists the plan at four times the access of its free tier, with expanded use of Gemini 3.1 Pro and Deep Research, its multi-source reporting tool.

Google's research tools deserve a specific mention, because they lean on the asset nobody else can copy: its index of the web. Deep Research, which Google describes as browsing and analysing hundreds of websites to assemble a report, is expanded on AI Pro, and the plan also widens access to Gemini Notebook, the note-and-source workspace that used to sit under the NotebookLM name. For a student or analyst whose raw material is the open web, that pairing is a genuine argument, not bundle padding.

That makes the value calculation unlike the other two. If you already pay Google for storage, a large part of the AI Pro price may already be in your budget, and the assistant arrives as an upgrade to tools you open every day. If you live outside Google's apps, the bundle is worth less to you, however capable the model is. One recent change sharpens the point: Google has been tightening which models the free Gemini app offers, reserving its Pro model for paying AI Pro and Ultra subscribers, so the gap between free and paid Gemini is wider than it was a few months ago.

Three questions, then the fine print

Benchmark charts invite the wrong argument. The frontier models trade places release by release, and a lead measured in March can be gone by October; anyone who has followed the open-source challengers closing on the frontier has watched that happen in public. For a paying individual, three practical questions sort the choice faster than any leaderboard.

Where does your work live? Inside Google's apps and storage, AI Pro meets you there. In a browser tab with a pile of PDFs and drafts, Claude Pro's Projects and long-document handling fit. Spread across voice, images, files and general tasks, ChatGPT Plus has the fewest missing pieces.

What do you hit first, limits or features? Heavy daily users run into usage windows on all three; Anthropic is simply the most explicit about how its five-hour and weekly limits work. Occasional users hit feature walls instead — no image generation here, no deep research there — and should pick on features.

And do you need more than one? Some professionals do pay for two, typically a writer's tool beside a research tool. Almost nobody needs three, and the overlap between any two is large. A useful discipline: run one subscription for a month with your real workload, note exactly what it refused or failed at, and let that list — not a launch event — decide whether a second subscription earns its place. The same scepticism we applied to Google's AI media detector applies here: a tool is what it demonstrably does, not what its name promises.

Three shared limits are worth knowing before any of them takes your money. None of these subscriptions includes API access: OpenAI states it plainly, Anthropic bills its Console usage separately from Pro, and Google's developer tools run on their own credits. A subscription is a seat in an app, not a balance for building software. Anyone budgeting for a product should price the APIs on their own pages and treat the consumer plans as a separate, personal expense.

Privacy terms also sit outside the headline price. OpenAI's help pages explain that conversations may be used to improve its models, with an opt-out available in the data controls. Anthropic's pricing page lists model training as opt-out across its consumer plans as well. The exact wording differs by company and changes over time, so the setting is worth checking on whichever service you choose — particularly before pasting client work, contracts or anything else that is not yours to share. None of the three consumer plans should be treated as a confidential enclave; that level of guarantee lives in their business and enterprise tiers, at different prices and under different terms.

Support is the third quiet difference. A consumer subscription buys you a help centre and a billing page, not an account manager. When a rollout changes the models on your account mid-month — and all three companies adjust availability as they ship — the individual plans simply absorb the change. Businesses that cannot absorb it are, again, pointed at the team and enterprise tiers.

People hesitate over these subscriptions as if they were phone contracts. They are not. All three bill monthly, all three can be cancelled from a settings page, and your chats and files stay attached to the account if you drop back to a free plan. The real switching cost is habit: the prompts you have tuned, the projects you have organised, the history a service has built up about how you work. That cost grows the longer you stay, which argues for choosing deliberately at the start — and for exporting anything genuinely valuable rather than leaving the only copy inside one company's app.

It also argues against subscribing in a panic during a busy week. Limits feel very different under deadline pressure, and a plan bought to escape one bad afternoon may sit unused for the other twenty-nine days of the month. A calmer test: for one week, keep a rough note of every time a free plan refuses you, slows you down or lacks a tool. If the note is short, stay free. If it fills a page, it is also your shopping list — it names the exact wall you need a subscription to remove, and usually points at one clear winner among the three.

One last thing, because the marketing will not say it. All three free plans are genuinely usable for light work — a few questions a day, short drafts, the occasional summary. The paid tiers earn their money through volume, speed, the strongest models and the extra tools. If you are not yet bumping into a free plan's ceiling at least once a week, you are probably not the customer these subscriptions were priced for. Wait until the ceiling annoys you. Then you will know which ceiling it was, and the choice above will make itself.