$500 for ChatGPT. Half the allowance per dollar.

OpenAI has launched a $500-a-month ChatGPT subscription. It comes with 25x the Plus allowance. The previous $200 plan offered 20x. That's 2.5 times the price for 25% more allowance: half the included allowance per dollar compared with the old deal.

Meanwhile, the $200 plan's Work and Codex allowance is reportedly being halved. Pay much more for a modest increase over what I had, or keep paying $200 for less. That's a pretty bullshit choice.

I moved my main coding workflow from Claude Code to Codex over the past month. I've really enjoyed it. Now I'm planning to move back to Anthropic, and the frustrating part is how much I still like the product I'm leaving.

What $500 actually buys

OpenAI introduced Pro 500 on September 29, alongside reopening Pro 200 to new subscribers. Here's the allowance comparison, using the reported former and revised Pro 200 figures and OpenAI's announced Pro 500 figure:

The final column is my arithmetic. These are allowance multipliers, not token balances or a promise of a fixed number of finished projects. Pro 500 also adds Astra Ultrafast access, which may matter to you. The 50% comparison is about included allowance per dollar, not every feature in the subscription.

OpenAI's Pro tiers page confirms the lower allowance at the unchanged $200 price; the exact 20x-to-10x cut comes from reporting.

Already on Pro 200? Check your grandfathered allowance before cancelling.

Eligible existing subscribers keep their full previous allowance through October 29, 2026, as long as Pro 200 stays active. If you're planning to leave, you can still use that allowance during the transition. Scheduling cancellation preserves access only until your paid billing period ends, so check that date and your eligibility in your account. The lower allowance starts October 30.

The Next Web also reports a one-time transition credit. That may soften the change for eligible users; it doesn't restore the ongoing allowance I'm deciding whether to pay for.

I'd have welcomed an extra tier that gave heavy users more room. Reducing my current plan while offering half its former allowance per dollar on the new one is hard to get excited about.

Why I paid $200 in the first place

The advertised comparison I was looking at when I switched was simple: $100 for 5x usage, $200 for 20x.

Twice the price bought four times the advertised allowance. Paying more gave me a better deal on the capacity I actually used. That made sense.

There is a caveat: OpenAI's April release notes describe temporary usage promotions. The earlier offers weren't a guarantee of permanent capacity. But the advertised allowance was part of what made me switch, and it shaped whether I felt the subscription was worth paying for.

I usually exhausted my weekly allowance after about four days. Then I'd take a break or move to another tool. My subscription was already doing its bit for work-life balance.

Even with those interruptions, I liked the deal. Reducing that allowance makes staying much harder to justify. It doesn't prove I'll now run out in exactly two days; tasks and settings vary. It does mean the margin I already needed is shrinking.

I still love building with Astra

GPT-6 Astra is my favourite model to build with so far. It understands visual material well, follows my instructions, and helps me get things looking finished. I've been building things with it that I'm excited to show people.

OpenAI's image generation is another reason I like the ecosystem. That's a separate capability from Astra's visual understanding, and I value both. A pricing announcement doesn't undo any of that.

That's why this annoys me. I'd happily keep using a product I enjoy. I just need enough access to it for the subscription to make sense.

More efficient models could help. Show me in my work.

OpenAI ties the lower allowance to improving model efficiency. I understand the argument: if a task consumes less of your budget, a smaller budget might still finish as much work.

That's possible. It needs to hold for the work I'm actually doing, at the quality I want. My decision to switch was heavily based on Astra. Moving to a cheaper model doesn't automatically preserve what I liked about that workflow.

I'd judge the change on completed work, corrections and time spent using comparable tasks and settings. Half the reported allowance isn't automatically half the output. It also isn't automatically harmless because another model got cheaper.

I'm moving back to Claude Code

Anthropic released Opus 5.5 on September 22, with a strong pitch on performance and efficiency. Those are vendor claims, and I'll judge it on my own projects. I don't need it to win every benchmark to reconsider where my money goes.

Claude has usage limits too. I'm not expecting unlimited compute or assuming its plan multipliers are equivalent to OpenAI's. I want a subscription that fits a working week without making me regret choosing the model I actually prefer.

Unless OpenAI changes the value on offer, my plan is to cancel Pro 200 after the transition and move my main coding work back to Claude Code.

It's a shame. OpenAI made something good enough to get me to switch. I'd have liked the subscription to give me a reason to stay.

Sources linked throughout. Checked September 30, 2026. Prices are published USD prices. Pro 500's 25x is from OpenAI; Pro 200's 20x-to-10x comparison is from The Next Web. Allowance-per-dollar figures are my calculations. Historical multipliers included promotions. My preferences and four-day usage pattern are personal experience.