Claude vs GPT is usually argued with benchmark charts and anecdotes. Every page on this shelf argues it with builds instead. Claude Opus 5.5 and Claude Fable 5.1 ran inside Claude Code; GPT-6 Astra and GPT-6.1 Sol ran inside OpenAI's Codex CLI. All four got the same prompts word for word, one attempt each, a 150-minute cap, and nobody touched the output. The rules are on the methodology page.
Three patterns show up before you read a single comparison:
- Time and volume. On the first four briefs the Claude models took roughly 3 to 11 times as long as the GPT models and wrote roughly 2.5 to 9 times as many output tokens.
- Variety. The two GPT models often converge on one idea: the same terracotta landing page, the same name for the physics demo, near-identical shooters. The two Claude models diverge more.
- Finishing. The longest run of all, Opus 5.5 on the shooter, hit the time limit and was published as it stood. That is on the record, not hidden.
Use the pages below when you want a question answered for a specific job: coding in general, a website, a decision between subscriptions. Use the model pairs when you want two specific models on every brief at once, and the briefs when you would rather skip the reading and watch.