ChatGPT 5.6 Sol, Terra and Luna: The Bill for Code Generation

Sol, Terra and Luna: The Real Cost of Programming with ChatGPT 5.6

Does it still make sense to pay €100 a month for Claude's top plan when ChatGPT costs €22 and Claude Pro costs €20? This question has been debated for over two months among those who earn their living writing code, and no one has settled it. At the heart of the issue is ChatGPT 5.6's model ladder—Sol, Terra, and Luna—with three levels of effort: minimum, high, and ultra. Terra, the intermediate level, has received the worst press: it costs more than the other two and doesn't always get it right on the first try.

Sol at Minimum or Luna at High: Where Spending is Decided

The positions fall into two camps. One argues that for most common usage, ChatGPT 5.6 Sol in chat mode with minimum effort is sufficient, and that the ultra variant—with parallel agents deployed—gets the job done but consumes tokens everywhere and requires more time investment than it saves. After over 25 hours of use, that recommendation holds: Sol at minimum is more than enough.

Opposing this view, another group prefers Luna at high effort for writing code and dismisses Terra: intermediate in capability, expensive in consumption, and equally prone to failure as the extremes. A third approach defends that being able to connect a mobile phone to the computer running the project—programming, launching scripts, and analyzing issues from the phone—justifies €18 or €100 a month, depending on one's ambition.

Why the Same Tool Costs €100 or €4 a Day

There is one case that summarizes the problem: one hundred euros spent with Sol and a couple of response failures. The other side countered that this is simply not knowing how to use it, and there is some truth to that, because the bill doesn't just depend on the chosen model. Lowering the temperature to low and working with short contexts brought the expense down to €3 or €4 per day; with long contexts, consumption skyrockets.

It is the detail that almost no one considers when comparing subscriptions: two people using the same plan can spend twenty times more depending on how they phrase the request. The fine breakdown of which model and effort level combination is cheapest per task is exactly what should be available before making a decision.

GPT6 Astra, Kimi and the Rest of the Toolkit

GPT6 Astra garners the most enthusiastic praise: it generates functional WordPress plugins, multi-page conversion funnels in a single attempt, and extremely long prompts that allow other tools to build an entire website successfully. It also faces criticism: it gets carried away and delivers results without asking first.

In parallel, Kimi has earned its place by generating a graphic flow diagram from C code without intermediate tools, although some warn that credits run out very quickly. Claude completes the picture: two alternating €20 accounts to start projects and then pass the baton to Codex, with Opus 5 Extra in low gear, according to its defenders.

The result is a decision that should be made with a spreadsheet but is decided through faith. The numbers are on the table, so are the alternatives, and the €100 plan continues to be paid while excuses for cancellation are sought. No one has yet said what it would take to stop doubting.