After the latest Codex /goal mode came out, I realized that GPT 5.5 is really crushing Claude Code’s Opus 4.7...

It actually ran for a full night, 7 hours straight, without any human involvement...
And it only used about 20% of the weekly limit of the 5x Pro...
The only things I gave it were a basic overview of my entry model’s core concepts... and a few examples.
Before, CC just helped me write a backtesting module, and then randomly came up with 4-5 sets of combinations...
This time, for Codex /goal, I gave it all the possible directions for parameter tuning and combinations, so it could first fix the backtesting module itself, and then go research how to truly optimize.
What kind of volume increase counts as a volume expansion, what kind of delta reversal conditions count as a confirmed reversal, what kind of TP, SL, and BE are the best...
Before I went to sleep, I started the live run... and when I got up, it had already done the strategy research and started writing the module to connect to live trading...
I’ll check later how its combined-strategy research turns out...