I spent way too long today chasing a tiny bug that turned out to be one missing character đ
. Nothing unusual there. What caught my attention was how differently I behaved once I switched to Claude Fable 5 inside OpenGradient Chat.
Normally, when I'm debugging something that actually belongs to a real project, I start deleting parts of the code before pasting it into an AI chat. Variable names, comments, API routes... I end up spending almost as much time sanitizing the prompt as solving the problem.
Today I didn't.
I already knew Fable 5 had been posting strong coding numbersâ95.0 on SWE-bench Verified, 80 on SWE-bench Pro, 84.3 on Terminal-Benchâso I expected solid answers. The surprise wasn't the benchmark. It was realizing I finally stopped thinking about what I had to hide before asking for help.
The responses weren't magically perfect. I still had to push back on a couple of suggestions and test everything myself. That's just how coding works.
What felt different was the workflow. My attention stayed on the bug instead of constantly asking myself, "Should I remove this snippet first?"
It's funny how people compare AI models almost entirely by benchmark scores. After today, I'm starting to think the bigger productivity gain comes from not breaking your own flow every five minutes just because you're worried about where your code ends up
@OpenGradient #opg $OPG
Normally, when I'm debugging something that actually belongs to a real project, I start deleting parts of the code before pasting it into an AI chat. Variable names, comments, API routes... I end up spending almost as much time sanitizing the prompt as solving the problem.
Today I didn't.
I already knew Fable 5 had been posting strong coding numbersâ95.0 on SWE-bench Verified, 80 on SWE-bench Pro, 84.3 on Terminal-Benchâso I expected solid answers. The surprise wasn't the benchmark. It was realizing I finally stopped thinking about what I had to hide before asking for help.
The responses weren't magically perfect. I still had to push back on a couple of suggestions and test everything myself. That's just how coding works.
What felt different was the workflow. My attention stayed on the bug instead of constantly asking myself, "Should I remove this snippet first?"
It's funny how people compare AI models almost entirely by benchmark scores. After today, I'm starting to think the bigger productivity gain comes from not breaking your own flow every five minutes just because you're worried about where your code ends up
@OpenGradient #opg $OPG