Twenty dollars a month buys about two flat whites in most cities. OpenAI would like you to spend it on ChatGPT Plus instead, and the pitch is straightforward: pay up and the walls stop appearing. That’s roughly true, but the interesting part is which walls, and whether you personally keep bumping into them.
I’ve been on and off the subscription for over a year, mostly on, and the honest answer is that Plus is worth it for maybe half the people who buy it. Here’s what’s actually inside.
What ChatGPT Plus actually is
Plus is the same chatbot your free account gives you, with the throttle loosened and a few extra rooms unlocked. Same interface, same conversation history, same tendency to confidently invent a citation. The subscription doesn’t make the model smarter in any fundamental way. It gives you more of the good stuff per day, and access to tools that the free tier either caps hard or withholds entirely.
The price has held at $20/month in the US since early 2023, which is unusual for software. What’s changed is the value packed behind that number. Two years ago Plus bought you GPT-4 and not much else. Now it’s a bundle of half a dozen separate products wearing the same chat box.
Free versus Plus: where the difference bites
The gap isn’t one big feature, it’s a dozen small ceilings. If you use ChatGPT for a five-minute question once a day, you’ll never notice. If you’re doing actual work in it, you’ll hit the ceiling by mid-morning.
- Message limits. Free users get a modest number of messages on the top models before being dropped to something weaker. Plus gives you a much higher allowance, and the fallback when you do hit the cap is gentler.
- Reasoning models. The slower, thinking-before-answering models are the real quality jump for maths, code and multi-step planning. Plus gets far more of them per day.
- Deep Research. The agentic mode that spends 5–30 minutes browsing dozens of pages and returns a cited report. Free accounts get a token taste; Plus gets a workable monthly allowance.
- Video and image generation. Sora access sits behind the paid tier, and image generation limits are meaningfully higher.
- Projects, tasks and custom GPTs. Saved instructions, file folders per project, scheduled tasks, and the ability to build your own bots. Free users get a fraction of this.
- Priority when servers strain. Less dramatic than it sounds, but on a bad day it’s the difference between working and refreshing.
The features most subscribers pay for and then forget
Deep Research
This is the single most underused thing in the subscription. Ask it to compare five vendors on security posture and pricing, walk away for twelve minutes, come back to a structured report with sources you can check. It’s not flawless, and it will occasionally cite a page it clearly skimmed rather than read. But for competitive scans, literature reviews or “summarise everything published about X in the last year,” it replaces an afternoon.
Agent mode
ChatGPT can now drive a browser, click through a booking flow, fill a form, pull data off a page. It works maybe 70% of the time on simple tasks and gets confused by anything with a captcha or an unusual layout. That failure rate is fine for low-stakes chores and infuriating for anything you actually need done. There’s a good breakdown of where this class of tool still trips over itself in this look at agentic AI in practice.
Advanced Voice
Talking to it while walking the dog feels like a gimmick for a week and then quietly becomes how you plan things. The screen-sharing and camera features are the useful half: point your phone at a broken dishwasher error panel or a spreadsheet and ask what’s wrong.
Where Plus genuinely earns its keep
The workflows that justify $20 tend to share one property: they involve files or repetition.
Dropping a 40-page contract into a conversation and asking for every clause that shifts liability is a ten-second job that would take an hour manually. Feeding it a messy CSV of six months of expenses and asking for a category breakdown produces working Python and a chart. Uploading three versions of a proposal and asking what changed between draft two and draft three is the kind of tedious comparison nobody wants to do by hand.
None of that is possible, or at least not repeatedly, on the free tier. The file limits and message caps kill the workflow before it starts. If you want a clearer picture of how the underlying models behave across these tasks, this breakdown of GPT-4’s real-world strengths and weak spots is worth twenty minutes.
Where it still lets you down
Paying doesn’t buy accuracy. It buys capacity. The model will still invent a plausible-looking statistic with total confidence, still lose track of an instruction buried 30,000 words deep in a document, still render garbled lettering on any generated image containing words. Complex arithmetic inside long chains of reasoning can drift. Ask it to compare two legal clauses and it may quietly merge them.
The practical response is verification, not avoidance. Anything that will be published, sent to a client, or acted on financially gets checked by a human. That’s not a knock on Plus specifically; it’s just the current state of the technology. A useful overview of what has actually improved versus what remains stubbornly broken lives in this assessment of ChatGPT and OpenAI in 2025.
Getting more out of the subscription
Most people use maybe 20% of what they’re paying for, largely because they ask vague questions and get vague answers. “Write me a blog post about pricing” gets you mush. “Write 400 words on why annual billing discounts backfire for SaaS, in the voice of a sceptical CFO, no bullet points” gets something usable on the first pass.
The prompt itself is the product. A specific role, a word count, an audience, a format, an example of the tone you want, an instruction about what to leave out. If your results feel generic, the fix is almost always in the request rather than the model, and this guide to escaping generic answers covers the mechanics properly.
Also worth doing: set up Projects for recurring work. One project for a client, with their brand notes and past documents loaded, means you stop re-explaining context every Monday morning.
The honest maths on $20
If you bill your time at anything above $30 an hour, Plus pays for itself the first month you use it to compress a two-hour task into twenty minutes. That’s a low bar. The people who should skip it are those who ask three questions a week, or who mainly want image generation, where the free allowances plus a dedicated tool often work out cheaper.
Worth knowing: Claude Pro and Gemini Advanced sit at the same $20 price point, and each is stronger on particular tasks. Claude tends to hold up better on long-form writing and careful document reading. Gemini’s integration with Google Workspace is genuinely convenient if your whole life is in Gmail and Docs. Perplexity is a better pure research engine. None of them are so far ahead that switching transforms your output.
The practical move is a single paid month. Use it hard, run the workflows you actually care about, note how often you hit a limit that matters. If you go three days without opening it, cancel. If you find yourself opening it before your email, keep it.