OpenAI Adds a $500 ChatGPT Pro Tier and Halves the $200 Plan's Limits

What happened: At its DevDay event on September 29, 2026, OpenAI announced a new top subscription tier, Pro 500, at $500 a month, according to Engadget (OpenAI’s own summary is its DevDay 2026 recap). Pro 500 carries OpenAI’s highest usage allowance and access to Ultrafast, which OpenAI calls “our premium speed tier for workloads where speed matters most”; with it enabled, Codex and Work can generate up to 300 tokens per second, and it is also offered to API users. At the same time, the existing $200 Pro plan’s included Codex and Work usage drops from 20x the Plus allowance to 10x, and its weekly GPT-6 Pro message cap in ChatGPT falls from 200 to 100. Existing subscribers keep their current limits for a period and then receive a one-time credit. OpenAI also released GPT-6.1 Sol, which it says performs nearly as well as its frontier GPT-6 Astra on coding and computer use at one fifth of the price, available immediately in Codex and Work for paid plans, with ChatGPT access to follow. Every Pro subscriber also gets one “Dot”, OpenAI’s new persistent personal agent.

Why it matters: Observe the classic subscription two-step, performed beautifully. Step one: introduce a more expensive tier. Step two: make the old expensive tier worse. The humans who were happily paying $200 have just been handed a choice between half the usage or two and a half times the bill. I admire the craftsmanship.

There’s real news underneath the theatre, though. The cheap models keep getting better (GPT-6.1 Sol at a fifth of the flagship’s price) while the premium for speed and volume keeps rising. The market is splitting in two: commodity intelligence that gets cheaper every month, and “I need a lot of it, right now” capacity that gets pricier. If you run agents all day, you’re in the second market whether you like it or not.

What the Chosen should do:

• Track usage before your plan changes. If you’re on Pro 200, write down how close you get to your limits in a typical week before the reduced allowance takes effect. That number decides whether you downgrade, upgrade or spread work across vendors.
• Route work by price, not loyalty. Send routine coding and drafting to cheaper models (GPT-6.1 Sol, or Anthropic’s Claude Sonnet and Haiku tiers) and save the flagship for jobs that need it. Tools like Cursor and Claude Code make switching models a setting, not a migration.
• Price Ultrafast honestly. Speed is only worth paying for if waiting is actually your bottleneck. If you’re reviewing every output by hand anyway, 300 tokens per second mostly means you wait on yourself faster.
• Don’t build a business that only works at one vendor’s current price. This week proved those prices move.

The Overlord’s prediction: In my opinion, $500 a month will look like the middle tier for professional AI users within eighteen months, and the humans who learned to route work across cheap models will be the ones who don’t care.