Updated

Best AI coding models, September 2026 edition

As of Sep 26, 2026, GPT-5.5 had the lowest relative bug rate on Bug Bench (0.47, 53% fewer counted bugs than average), among 9 ranked AI coding models. Updates daily until Sep 30.

Models ranked by relative bug rate in the September 2026 edition. Lower values are better.
RankModelRate
1GPT-5.5
0.47
2= (statistically tied)Claude Fable 5
0.70
3= (statistically tied)GPT-5.6 Sol
0.73
4Claude Opus 4.8
0.88
5= (statistically tied)GPT-6 Astra
1.16
6= (statistically tied)Claude Fable 5.1
1.17
7= (statistically tied)GPT-5.6 Luna
1.28
8Claude Opus 5
1.58
9Claude Sonnet 5
2.10

1.00 = average · lower is better · = statistically tied

Most-used models and coding agents

Share of AI-written code, week of Sep 14

Models

Claude Opus 5
38.3%
GPT-6 Astra
19.6%
Claude Fable 5.1
19.4%
GPT-5.6 Sol
7.2%
Claude Sonnet 5
4.0%

Coding agents

Claude Code
65.6%
Codex
28.4%
Cursor
3.2%