TL;DR: OpenAI just launched a $100/month plan. Between Plus at $20 and Pro at $200, targeted at Codex developers, with five times more usage. We are building, without saying it, a two-tier AI market. On one side, free and cheap plans with throttled models. On the other, premium plans where AI is genuinely transformative. The problem: the public debate is based on the first. And that distorts everything.

In my two previous articles, I argued that the SDLC is dead and that coding was never the job. I was talking about process and roles. But there's a far more uncomfortable topic — and nobody wants to look at it.

When OpenAI announces a $100/month plan, wedged between Plus at $20 and Pro at $200, it's not a pricing adjustment. It's a confession. The confession of something I've been observing for months, and that the public debate refuses to name: we're not all using the same AI. And everything said about it is rigged from the start.

Two AIs, one name

Look at OpenAI's pricing grid today. Free. Plus at $20. The new plan at $100. Pro at $200. Four tiers. Four radically different experiences. (Sources: openai.com/chatgpt/pricing, openai.com/index/introducing-chatgpt-pro.)

On the free plan, you get a throttled model, tight quotas, degraded response times during peak hours, and limited access to the latest models. On Plus at $20, it's better — but quotas remain constraining, especially if you're doing serious work. You hit the ceiling by mid-afternoon. You wait. You rephrase to save tokens. You adapt your workflow to the tool's limits instead of the other way around.

On the $100 plan, five times more capacity. Priority access to Codex and the most powerful models. On Pro at $200, near-unlimited usage of the best models, deep research mode, extended capabilities.

Same brand. Same interface. Same name. But these are not the same products.

"This isn't the same AI at two prices. It's two AIs under the same name. One gives you an impression. The other reshapes your career."

And OpenAI isn't alone. Anthropic has its Max offering at $100 and $200 for Claude. Google pushes Gemini Advanced. Everyone is building the same structure: a free showcase for the masses, and a real product for those who pay.

The experience gap

This isn't just about quotas. It's about qualitative experience.

The person on the free plan interacts with a model that hallucinates more often, has less context, refuses more requests, and produces more generic answers. They reach the logical conclusion: "AI is impressive but unreliable."

The person on the $200 plan uses the same model name — but with extended context, deep reasoning capability, access to advanced tools, and quotas that enable a real continuous workflow. They reach a very different conclusion: "AI changed how I work."

Both people talk about "ChatGPT". They talk about "Claude". They think they're talking about the same thing. They're not talking about the same thing.

It's like someone judging photography after using a disposable camera, while another works with a Hasselblad. They both talk about "taking photos". But one captures blurry memories and the other produces art.

"The gap between free and premium AI isn't a performance difference. It's an era gap."

The public debate is biased at the source

And this is where it becomes a real problem.

The public debate on AI — the one that shapes regulations, corporate policies, investment decisions, public opinion — is overwhelmingly based on the free or cheap AI experience.

Journalists covering AI almost all use the free plan or Plus at $20. They test ChatGPT, notice hallucinations, write an article about "AI's limitations". Their conclusion is honest — but it's based on a product that's throttled on purpose. It's like test-driving a Ferrari with the handbrake on, concluding that sports cars are overrated, and selling the conclusion on the front page.

Regulators have seen a demo. Maybe two. Probably on a free account, probably prepared by an assistant who doesn't use AI daily. They legislate on a technology they've never used in a real workflow for a full week. The EU AI Act is a massive text. How many of its drafters used a Pro plan for a month to understand what the technology actually does?

Politicians have heard that "ChatGPT hallucinates". That AI generates false content. That it's dangerous. And they're not wrong — on the free plan, it often does. But nobody showed them what an advanced reasoning model does with a million-token context on a structured problem.

Business leaders deciding their organization's AI strategy often have the same limited experience. They tried ChatGPT on a Saturday afternoon, asked it to write an email, found it "nice but not revolutionary", and decided AI could wait until 2027 on the roadmap.

"We legislate the future of AI with the experience of a free account. It's like flying a fighter jet after flying a kite. Both leave the ground. They're not the same war."

Those who know don't talk

Meanwhile, a silent minority lives in a different technological reality.

The developer paying $100 or $200/month doesn't waste time debating on LinkedIn whether AI is useful. They're too busy delivering in a morning what used to take a week. They don't comment on "will AI replace developers?" articles — they're living the answer.

The researcher using Claude Pro with a million-token context doesn't debate whether AI hallucinates. They've learned to structure their queries, verify outputs, and use the tool in a workflow that produces reliable results. Hallucinations exist — but in a structured context with an advanced model, they're manageable, identifiable, and increasingly rare.

There is an asymmetric silence. Those with the deepest AI experience are the ones who talk about it least publicly. They have nothing to prove. They don't need to convince anyone. They work. And the gap between their reality and the public debate widens every month.

"Silence isn't neutral. Those who live the real product work. Those who only know the showcase comment. Guess which of the two groups shapes public opinion."

What this actually means

This segmentation isn't an accident. It's a perfectly rational business model. Free plans serve as a showcase and acquisition pipeline. Premium plans generate revenue. Every AI company does the same thing — and they have every right to.

But the consequences run deep:

1. We're creating a cognitive divide. Two groups of people discuss the same topic with fundamentally different experiences. One says "AI is a gimmick". The other says "AI transformed my work". Both are right — but they're not talking about the same product. And this misunderstanding contaminates every conversation: in business, in politics, in media.

2. Strategic decisions are made on the wrong basis. When a CTO decides their company's AI strategy based on their free-plan experience, they massively underestimate the technology. When a regulator legislates after a demo on a standard account, they regulate a product that doesn't exist one floor up. Decisions are made on faith in a truncated version of reality.

3. The productivity gap will accelerate. Companies and individuals who invest in premium plans build workflows, habits, and expertise that compound. Every month of intensive usage widens the gap. In a year, the difference between someone using free AI and someone on a Pro plan won't be 5x — it'll be an order of magnitude. Because it's not just the tool that's better. It's the person who learned to use it at full potential.

4. The AI debate is a class debate that won't say its name. $200/month is nothing for a California tech company. It's a significant budget for a freelancer in France. $100/month for Codex is a reasonable investment for a well-paid senior developer. It's unreachable for a student or junior developer. We're building a technology where real access — not marketing access, real access — correlates with purchasing power. This isn't new in tech. But the gap has never been this wide between the showcase and the product.

What should be done

I'm not naive. AI companies need revenue to fund colossal infrastructure costs. Tiered pricing is legitimate. I'm not asking for free access for everyone.

But I'm asking that we stop pretending the debate is honest.

For journalists: if you cover AI, you need to use premium plans. Not once. Daily, for weeks. Otherwise you're testing a marketing product, not the technology. It's like covering motorsport by driving a compact car. Your newsroom should budget a Pro subscription like it budgets a Reuters feed.

For regulators: before legislating, use the technology. Not a demo. Real usage, on a real workflow, for a real month. Put a Pro plan on your office's budget. It's $200 — less than a working lunch in Brussels. If you regulate the future of AI without touching the real product, you're regulating the future of aviation after a glider flight.

For business leaders: don't make strategic AI decisions based on your personal free-plan experience. Give premium plans to your best people for three months. Measure results. Then decide. The cost of a Pro plan for an entire team is less than the cost of a consultant who'll come explain what AI can do — poorly, since they probably don't use it on the Pro plan either.

For individuals: if AI is part of your job, invest in a premium plan. It's a professional tool. A developer who codes on a free plan and concludes AI is "mediocre" is a photographer judging their art with a disposable camera. You can't evaluate a technology you've never used at full potential.

The real divide

The AI divide isn't between those who use AI and those who don't. That would be too simple.

"The real divide is between those who use premium AI and those who think they're using the same thing. The first group ships. The second comments."

The free plan gives an impression. The premium plan gives a transformation. Between the two, there isn't a quota difference. There's a change of species.

And as long as the public debate is fed by people who've only known the impression, the decisions that follow will be disconnected from reality. We'll regulate yesterday's technology. We'll invest too little, too late. We'll lose ground to competitors who understood that $200/month isn't a subscription — it's a competitive advantage.

"The real danger in 2026 isn't that AI is overestimated. It's that the people underestimating it are deciding for the people living it. And they call it caution."

Sources

  • OpenAI — ChatGPT pricing: official page listing every ChatGPT plan (Free, Plus at $20/month, Business, Pro at $200/month) and their respective quotas.
  • OpenAI — Introducing ChatGPT Pro: official announcement of the ChatGPT Pro plan at $200/month, with expanded access to advanced reasoning models and deep research mode.
  • OpenAI — Codex: official page for the Codex development agent, bundled into the Plus, Business and Pro plans with differentiated quotas.
  • Anthropic — Claude pricing: official page listing the Claude plans (Free, Pro at $20/month, Max at $100 and $200/month) with their usage quotas and available models.
  • Google — Gemini subscriptions: official page for Gemini plans (Free, Gemini Advanced via Google AI Pro and Google AI Ultra) with the latest Gemini models and extended capabilities.

Read also: The SDLC Is Dead — why the process was always broken, and why AI agents are making it impossible to pretend otherwise. And its follow-up: Coding Was Never the Job — if AI killed the process, what happens to the roles?