Fable 5.2 incorrectly fixes bug in 3 out of 4 trials

· Blog Polybdenum · Nov. 17, 2026, 3:39 a.m.
Summary
The author evaluates the performance of Fable 5.2 compared to Claude Code (Opus 5.5) by testing a bug detection scenario in their project 1SubML. While both models identified correct fixes in a limited number of trials, the results suggest Fable 5.2 has not advanced significantly over Opus 5.5, leaving room for human oversight in coding. The author shares their unsatisfactory experience when seeking advice on improving their project with Fable, leading them to question its usefulness.