The plan finished baking. Reviewed, revised, the whole nine yards. I wasn’t going to execute the plan myself; I wanted to automate it as well. The only decision left was how.

Claude Code is my go-to vibe coding AI, but the tasks at hand weren’t coding. Majority of them were browser control. Considering how slow AI-driven browser use is, Claude already had an issue. Its five hour session window would require me to resume the project constantly, not to mention all the lost tokens picking up the work. The only option to make it work with Claude would have been to pay for usage credits, and going by community reports and third-party benchmarks, subscriptions are heavily subsidized against credits.

In a way, I was lucky OpenAI has lifted the rolling window system on the Codex. In July, with the release of GPT 5.6, OpenAI temporarily lifted the system. I was able to use this window of opportunity. The migration plan was already laid out. I just needed Codex to work continuously, uninterrupted. Had I not been able to choose an alternative, I would have resorted to finding an automator that periodically resumes itself after the window is open again.

The model I used on Codex, 5.6 Sol, is not free. To be specific, I used the $100 ChatGPT Pro subscription. A year of hosting and domain is roughly $80 originally, and recurring bill for domain only will be $10 per year. That $70 saved every year writes the subscription off in under two, and I can use Codex for other tasks, so the price was more justifiable than paying $80 a year for one blog.

As for Sol itself, the model met expectations. The recommendation came from Claude Code user communities. As part of an effort to keep token usage low and continue working past the five hour window, Sol was often recommended as a cheaper replacement than Fable, with good reviewing capability. I could see how Sol earned its reputation. Execution took 48 hours or so, and combined with the usage from planning and reviewing, Sol on high effort had only used half of the weekly usage on the $100 Pro plan.

However, this does raise some important nuance to prospective bloggers who are considering migrating using AI as well. If you are migrating, chances are the longer plan will drain the five hour window during execution. As it stands, the only solution presented by both platforms is to fall back to usage credits until the window resets. Neither Codex nor Claude Code particularly have transparent write ups on how usage credits fare against subscriptions. It’s a Pass on results, but a complete disastrous F on transparency at the moment.