OpenAI quietly updated their pricing page last week, and if you blinked you missed it. But the changes are significant — especially if you're on the free tier, a Plus subscriber considering Pro, or a developer using the API.
Here's what actually changed, who benefits, and who's going to pay more.
The New Tier Structure
- GPT-4o mini
- Limited messages
- Basic web browsing
- No image gen
- GPT-4o full access
- DALL-E image gen
- Advanced analysis
- Custom GPTs
- Unlimited GPT-4o
- GPT-5.4 access
- o3-pro reasoning
- Extended context
- All models
- SSO & admin
- Dedicated capacity
- SLA guarantee
What Changed
Free Tier: More Restrictive
The biggest change for casual users: the free tier is now limited to GPT-4o mini only. Previously, free users could access GPT-4o with message limits. That's gone. If you want GPT-4o, you need Plus.
OpenAI says this is about "ensuring quality of service for paying users." Translation: free users were overwhelming GPT-4o's capacity during peak hours.
Plus Tier: Unchanged (For Now)
The $20/month Plus tier stays the same — GPT-4o access, DALL-E, Advanced Data Analysis, and Custom GPTs. But there's a notable absence: GPT-5.4 is not included in Plus. You need Pro for that.
Pro Tier: The Real Story
The $200/month Pro tier is where things get interesting. Launched in late 2025, it was initially seen as a niche product for power users. But with GPT-5.4 exclusive to Pro and Enterprise, it's now the only way individuals can access OpenAI's most powerful model.
"We believe the best models should be accessible, but running frontier AI at scale requires sustainable pricing. Pro gives power users the capacity they need while keeping Plus affordable for everyone." — OpenAI Blog, March 2026
API Pricing: The Numbers
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Change |
|---|---|---|---|
| GPT-4o | $2.50 | $10.00 | ↓ 17% cheaper |
| GPT-4o mini | $0.15 | $0.60 | ↓ 25% cheaper |
| GPT-5.4 | $15.00 | $60.00 | New model |
| o3-pro | $20.00 | $80.00 | New model |
| GPT-4 (legacy) | $30.00 | $60.00 | Deprecated June 2026 |
The pattern is clear: older models get cheaper, frontier models carry a premium. GPT-4o and 4o mini are now the cheapest they've ever been. But if you want the cutting edge, you're paying frontier prices.
How It Compares to Anthropic & Google
| Provider | Consumer Plan | Frontier Model | API (input/1M) |
|---|---|---|---|
| OpenAI | $20 Plus / $200 Pro | GPT-5.4 (Pro only) | $15.00 |
| Anthropic | $20 Pro | Claude Opus 4.6 | $15.00 |
| $20 Gemini Advanced | Gemini 2.0 Ultra | $12.50 |
It depends on what you do. If you primarily use ChatGPT for writing, research, and general tasks, Plus is still plenty — GPT-4o handles those extremely well. Pro makes sense if you: (1) need GPT-5.4 for complex coding or reasoning tasks, (2) hit Plus rate limits regularly, (3) use o3-pro for math, science, or complex logic problems, or (4) need extended context windows for processing long documents. For most users, Plus remains the sweet spot. Pro is for power users who've already maxed out what Plus can do.
The Bigger Picture
OpenAI's pricing evolution tells a story about where the AI industry is headed: the base is getting cheaper, the frontier is getting more expensive.
This makes sense economically. Running GPT-4o has gotten cheaper as inference costs drop and hardware improves. But training and running GPT-5.4 — which is estimated to have cost over $500M to train — requires premium pricing to recoup investment.
For consumers, the practical impact is:
- Students and casual users: The free tier is weaker, but Plus at $20/month is still competitive. Get your free student access if eligible.
- Power users: You're going to need to decide between Pro ($200/mo) and exploring alternatives like Claude or Gemini for frontier performance at lower cost.
- Developers: API costs for standard models are dropping. Frontier API access is expensive but powerful. Multi-provider strategies make more sense than ever.
AI is splitting into two markets
Commodity AI (GPT-4o, Claude Sonnet, Gemini Flash) is getting cheaper by the quarter. Frontier AI (GPT-5.4, Mythos, Gemini Ultra) is getting more expensive and more restricted. Where you sit on that spectrum determines your costs — and your options.