OpenAI Is Paying Me to Move My Business to Anthropic

Three months ago, OpenAI was my daily driver, using Codex with GPT-5.5 and being productive like never before. Anthropic already had Fable but it wasn't really available through the subscription, at least not fully, so I was using it only for occasional reviews. Then OpenAI launched Luna, Terra, Sol, Astra with smaller updates every couple of weeks while Anthropic launched Fable 5.1, Opus 5.5 and Sonnet 5.5. The newer GPT models might be intelligent, but they are annoying to use and burn through my usage limits way too fast, while the new Claude models are finally good again at every task I give them. And now weirdly it looks like OpenAI is paying me to use more of Claude while killing the one thing that moved people from Claude to Codex which is generous usage limits in the 200 USD subscription.

OpenAI just gave me (and everyone else I guess) $5,000 in usage credits, split across my two Codex accounts. The same email says the usage allowance included in my $200 subscriptions will be cut in half. So I am going to downgrade both accounts to $20 Plus, use the credits, and put the money I save toward two more Claude accounts.

The $5,000 OpenAI usage credit lets me keep using Codex while I send OpenAI much less subscription money, and that makes it easier to move more of my business to Claude.

Right now I pay for two $200 Codex subscriptions and two $200 Claude subscriptions. That is $800 a month before API usage, and the cost of AI is already my main blocker. I have more things to build than I can afford to give to the models.

From October 30, 2026, Pro 200 will include 10x the Plus allowance for Codex instead of 20x, of course the price stays at $200.

The new lineup is Plus at $20, Pro 100 at $100 with 5x the Plus allowance, Pro 200 at $200 with 10x, and Pro 500 at $500 with 25x. That is the same amount per dollar all the way up. Pro 200 has been giving me twice as much nominal allowance per dollar as Plus, and that advantage ends on October 30. So there are no more volume discounts from now on which feels a bit like the beginning of the end of subsidised AI subscriptions. It's only a short way from here to full API pricing.

I have been cautious about OpenAI's newer models anyway. Since GPT-5.6 Sol, my problem with them has been overly long task runs. The models are smart, and the final result is good. But when I give Sol a task, it keeps verifying, adding adjacent work, reviewing its own reviews, and always finding more to do. Sometimes it runs for hours, sometimes for a day or two, and sometimes it uses up the whole usage limit.

I went back to GPT-5.5 after trying Astra. For ordinary tasks including coding, 5.5 would often do what I asked and come back after a few minutes, or push through independently when I ask for it. I could review the result and move on. With Sol and Astra, I always have to worry whether a session would ever finish.

I still use Luna. It is useful and cheap enough that paying for usage does not bother me much. I used to go to DeepSeek when my subscription ran out, and now Luna is my goto choice that replaces DeepSeek, especially as the new DeepSeek v4.1 flash is a lot more expensive than v4.

I do wonder whether it is the beginning of the end of the generous subscription plans that made using SOTA models affordable for indie devs, freelancers and small companies. OpenAI argues that more efficient and more capable models will still let me get more done. But my problem in the last couple of weeks has been getting those models to finish within the limits I already had.

What I'm building

Delegate tasks. Get software.

Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request.

Take a look at vroni.com