I Still Pay for ChatGPT — Here Is Why
Four years. That is how long I have been paying OpenAI $20 a month. I have cancelled plenty of subscriptions in that time — streaming services, meal kits, that meditation app I used twice. ChatGPT is still here.
But the gap is narrowing. Claude now produces better code. Gemini has better Google integration. DeepSeek is free and competitive on coding. So why am I still paying? Here is what I found after 3 months of testing ChatGPT against Claude and Gemini for daily work.
By the Numbers
ChatGPT is still the biggest AI product on earth. 1.1 billion monthly active users. 900 million weekly actives. $25 billion in annual revenue. That is more than twice what it made in 2024.
But here is what should worry OpenAI: in May 2026, ChatGPT’s market share dipped below 50% for the first time — sitting at 46.4%. Claude and Gemini are eating into their lead. The question is whether this is a blip or a trend.
OpenAI has about 50 million paying users. Roughly 10 million of those are on the $20/month Plus plan. The rest are on Team, Enterprise, or the $200 Pro tier. The company is now valued at $852 billion and is actively preparing for an IPO.
One thing that caught me off guard: ads. OpenAI is testing advertisements in the free tier. About 17% of daily active users have seen them so far. For now, paid users do not see ads. I would not count on that lasting forever — OpenAI needs to show IPO investors a path to profitability beyond subscriptions.
What $20 Actually Buys
The Plus plan removes most of the annoying restrictions from the free tier. You get more messages with GPT-5.6 before hitting the cap, access to the o5-mini and o4-mini reasoning models, DALL-E 4 for images, the Code Interpreter sandbox for data work, Advanced Voice, and the ability to build your own custom GPTs. Web search, file uploads, and conversation memory are included.
For $20, that is a lot. Two years ago, half of these features did not exist.
I use regular GPT-5.6 for about 80% of what I do in ChatGPT. It responds in a few seconds. The writing sounds natural. It follows conversation threads without losing context. For email drafts, document summaries, headline brainstorming, and quick questions, it does the job.
Where GPT-5.6 Still Stumbles
I do not trust it with facts without web search turned on. In March, I asked it to list the top 10 VC deals in Europe for Q4 2025. It gave me a confident list — names, amounts, rounds. Every single one was wrong. The companies existed, but the numbers were fabricated. This is not a one-off. I have hit similar hallucinations about 1 in 15 factual queries.
The o-series reasoning models help. They “think” before answering and produce more reliable results on math, logic, and technical questions. But the thinking is hidden — you see “Thinking…” for 10-60 seconds, then you get the answer. Claude shows you its reasoning, which I find more useful when I need to verify the logic.
Features That Actually Matter
Code Interpreter
This is the feature I would miss most if I switched. Upload a CSV, ask for analysis, get charts and stats. It runs Python in a sandbox with pandas, matplotlib, numpy, and scikit-learn pre-installed. I have used it to clean client data, prototype visualizations, and run quick statistical tests that would normally take me 20 minutes in Excel.
The sandbox is ephemeral — files disappear after your session. You cannot install arbitrary packages. It times out after a few minutes of inactivity. These are annoyances, not dealbreakers.
DALL-E 4
Built-in image generation is convenient. I use it for blog post headers, social media graphics, and quick mockups. The images are good enough for web use. Text rendering has improved — it used to produce gibberish, now it gets it right about 80% of the time.
If you need photorealistic images, use Midjourney. If you need precise control, use Adobe Firefly. DALL-E 4 is about speed and convenience, and it delivers on both.
Advanced Voice
I did not expect to use voice as much as I do. Driving to meetings, cooking dinner, walking the dog — I pull out my phone and talk to ChatGPT. The voice sounds natural. It handles interruptions well. Latency is under 1 second most of the time.
The main limit: you get fewer voice messages than text messages, even on paid plans. For me, it is about 30 minutes of voice per day on the Plus plan before it drops me to a less capable voice model. That has been enough, but heavy users will notice.
Custom GPTs
I built a custom GPT for my weekly newsletter workflow. It has my style guide, past issues, and formatting instructions loaded as knowledge files. It saves me about 45 minutes per issue. That alone justifies the $20.
The GPT Store is a mixed bag. Most custom GPTs are thin wrappers around GPT-5.6 with a system prompt. The good ones — like the one that has a company’s full documentation loaded — are genuinely useful. But you have to wade through a lot of noise to find them.
The $200 Pro Plan
For 95% of people, the answer is no — do not pay $200/month. Here is why.
I tested the Pro plan for 2 weeks. The main benefit is unlimited access to o5, OpenAI’s most advanced reasoning model. o5 is better than o5-mini on edge cases: complex legal analysis, multi-step physics problems, deep code review. But the gap is smaller than you would expect from the 10x price difference.
You also get unlimited GPT-5.6 usage. On the Plus plan, I hit the limit about twice a month — usually when doing long back-and-forth coding sessions. If that describes your daily workflow, Pro might make sense. If not, save the $180.
The Competition is Catching Up
ChatGPT’s edge in 2024 was that it did everything better than everyone. That is no longer true.
Claude produces better code and more careful analysis. Its 200K context window (vs ChatGPT’s 128K) matters for long documents. And Anthropic now leads ChatGPT in revenue per user — $2.76 vs $1.74 — which suggests users are willing to pay more for Claude’s quality.
Gemini has 662 million monthly users and the best Google integration. If you live in Google Workspace, Gemini saves you time every day in ways ChatGPT cannot match.
And then there is DeepSeek, which is completely free and competitive on coding.
ChatGPT’s defense is breadth. It does more things than anyone else. For users who want one tool to cover everything, ChatGPT is still the default. But more people are adopting a two-tool strategy — ChatGPT for everyday tasks plus Claude for serious analytical work. That is a risk for OpenAI’s subscription growth.
The Ads Situation
Here is something that does not get enough attention. OpenAI is testing ads on the free tier. About 17% of daily active free users see them. The company has said paid users will not see ads “at this time.” That qualifier — “at this time” — is doing a lot of work.
I understand why they need to do this. $25 billion in revenue sounds impressive until you compare it to their $852 billion valuation. The math only works if revenue keeps growing fast. Ads are the obvious next lever. But they change the user experience in ways that are hard to predict.
What I Actually Use ChatGPT For
After 3 months of testing all three major assistants, here is where ChatGPT still earns its place in my workflow:
-
Email drafts. GPT-5.6 writes natural, conversational emails faster than Claude or Gemini. I send about 40-50 work emails a week and ChatGPT handles first drafts for about half of them.
-
Data analysis. Code Interpreter is unique. When I need to explore a dataset, I upload it to ChatGPT. Claude can write the code but cannot run it. Gemini can run basic analysis in Sheets but lacks a proper sandbox.
-
Image generation. When I need a blog header or social graphic, DALL-E 4 is 10 seconds away. Not the best images, but the fastest.
-
Quick questions. For “what does this error mean” or “summarize this article” or “write a subject line for this email,” ChatGPT is fast and reliable enough.
What I Use Other Tools For
-
Coding and code review goes to Claude. Better code quality, better debugging, and Claude Code is a genuinely useful CLI tool.
-
Long document analysis goes to Claude for quality (200K context, better attention) or Gemini for size (1M token context).
-
Research goes to Gemini when I need current information or to NotebookLM when I need to synthesize multiple sources.
-
Deep reasoning on hard problems goes to Claude with extended thinking. I can see the reasoning and verify it.
Is ChatGPT Still Worth $20?
I will keep paying. Not because ChatGPT is the best at everything — it is not. But because for $20, no other single tool gives me this much utility. I can write emails, analyze spreadsheets, generate images, debug code, and talk to an AI while driving — all from one subscription.
The free tier is good enough for casual use if you can tolerate the message caps. The $200 Pro plan is hard to recommend unless you are doing deep research or heavy coding every single day.
If you can afford one AI subscription: ChatGPT Plus, no hesitation. If you can afford two: add Claude Pro for coding and analytical work. That $40/month combo covers 90% of what a knowledge worker needs from AI.
The three things I am watching: the ads creeping into the free tier, the IPO pressure on pricing, and whether Claude or Gemini can close the feature gap before OpenAI fixes its quality gap.