Updated
Best AI coding models, September 2026 edition
As of Sep 30, 2026, GPT-5.6 Sol had the lowest relative bug rate on Bug Bench (0.72, 28% fewer counted bugs than average), among 8 ranked AI coding models. Final September 2026 ranking.
| Rank | Model | Rate |
|---|---|---|
| 1 | GPT-5.6 Sol | 0.72 |
| 2 | Claude Opus 4.8 | 0.79 |
| 3= (statistically tied) | Claude Fable 5.1 | 1.21 |
| 4= (statistically tied) | GPT-6 Sol | 1.30 |
| 5= (statistically tied) | GPT-6 Astra | 1.30 |
| 6= (statistically tied) | Claude Opus 5 | 1.50 |
| 7 | Claude Opus 5.5 | 1.78 |
| 8 | Claude Sonnet 5 | 2.07 |
1.00 = average · lower is better · = statistically tied
Most-used models and coding agents
Share of AI-written code, week of Sep 21
Models
- Claude Opus 5.5
- 30.5%
- Claude Opus 5
- 23.3%
- Claude Fable 5.1
- 13.4%
- GPT-6 Astra
- 13.1%
- GPT-6 Sol
- 4.3%
Coding agents
- Claude Code
- 72.5%
- Codex
- 22.7%
- Cursor
- 3.0%