Which AI model writes the most bugs in Shell?

As of Sep 30, 2026, Claude Opus 5 had the highest relative bug rate in Shell code (1.40, 40% more counted bugs than the Shell average), statistically tied with 4 other models.

Updated Sep 30, 2026

Best AI model for Shell code

As of Sep 30, 2026, GPT-5.6 Sol had the lowest relative bug rate in Shell code (0.61, 39% fewer counted bugs than the Shell average), statistically tied with 5 other models, among 7 ranked models.

Models ranked by relative bug rate in Shell code. Lower values are better.
Rank	Model	Rate
1= (statistically tied)	GPT-5.6 Sol	0.61
2= (statistically tied)	Claude Opus 4.8	0.94
3= (statistically tied)	Claude Sonnet 5	1.06
4= (statistically tied)	Claude Fable 5.1	1.10
5= (statistically tied)	GPT-6 Astra	1.21
6= (statistically tied)	Claude Opus 5.5	1.23
7= (statistically tied)	Claude Opus 5	1.40

1.00 = Shell average · lower is better · = statistically tied

Other languages

TypeScript
JavaScript
Python
Go
Rust
SQL
Ruby
Swift
PHP
Vue
Terraform
Kotlin
C#
Svelte
Java
Dart

Questions and answers

Which AI model writes the most bugs in Shell?: As of Sep 30, 2026, Claude Opus 5 had the highest relative bug rate in Shell code (1.40, 40% more counted bugs than the Shell average), statistically tied with 4 other models.