Which AI Model for Which Job, Tested
Pick the job you have. Each page shows which model scored best on it, which one is cheapest for about the same score, and what made a real difference.
Each line names the top score on each test method. Scores from two test methods are never added together.
| Job | Top score with no help |
|---|---|
| Accessibility audit | Via API: each model was given tasks at its own level of difficulty, so no one model is ranked first; In Claude Code: each model was given tasks at its own level of difficulty, so no one model is ranked first. |
| Code review | Via API: GPT-5 mini 70.0%; In Claude Code: Claude Opus 5 93.5%. |
| SEO audit | Via API: Gemini 3.1 Flash Lite 48.8%; In Claude Code: Claude Fable 5.1 97.2%. |
| Writing a skill | Via API: Gemini 3.1 Flash Lite 87.3%; In Claude Code: Claude Fable 5.1 100.0%, Claude Opus 5 100.0%, Claude Sonnet 5 100.0% and Claude Fable 5 100.0%, tied. |
| Writing a spec | Via API: Claude Haiku 4.5 99.5%; In Claude Code: Claude Opus 5.5 99.5%. |
| Link and metadata audit | In Claude Code: Claude Fable 5.1 100.0%. |
| Data checking | In Claude Code: Claude Opus 5 75.8%. |
| Traffic drop triage | In Claude Code: Claude Opus 5 100.0% and Claude Fable 5.1 100.0%, tied. |
| Clean JSON | Via API: no model managed this on its own; In Claude Code: no model managed this on its own; In Codex: no model managed this on its own. |
| Exact length | Via API: GPT-5 mini 5.0%; In Claude Code: Claude Haiku 4.5 4.2%; In Codex: GPT-5.6 Terra 62.4%. |
| Recent events | Via API: Claude Haiku 4.5 85.6%; In Claude Code: Claude Haiku 4.5 96.7%; In Codex: only one model was measured on this run, so there is nothing to rank. |
| Step-by-step answers | Via API: Gemini 3.1 Flash Lite 40.0%; In Claude Code: Claude Fable 5 40.0%, Claude Fable 5.1 40.0%, Claude Opus 5 40.0%, Claude Sonnet 5 40.0% and Claude Opus 5.5 40.0%, tied; In Codex: GPT-6 Sol 40.0%. |
How these lines are read
A line names the model with the highest score on the job, on each test method that ran it. Where two or more could not be told apart, it names them all. The same picks, with the cheapest, the most consistent and the fastest, are on the routing page and in a data file.