← Back to home
llm comparison

GPT-4o vs Claude Sonnet 4: Real-World Test Results After 30 Days

We used GPT-4o and Claude Sonnet 4 daily for code writing, email drafting, research synthesis, and creative tasks. Here's what actually happened.

David Chen · 2024-10-17
Independent AI researcher and content creator. Tests AI tools daily across image generation, LLMs, and productivity workflows. Based in China, serving a global audience.

Background

Code Writing

TypeScript / Next.js

**GPT-4o:** Produced working code 85% of the time. Occasional hallucinations around Next.js App Router conventions. Faster, more confident.

**Claude Sonnet 4:** More conservative, often produced more idiomatic code. Better at explaining *why* something was wrong.

**Winner: GPT-4o** for speed and completion rate; **Claude Sonnet 4** for code quality and explanation.

Python / Data Analysis

**Claude Sonnet 4:** Noticeably better at pandas operations and understanding data pipeline logic. Fewer off-by-one errors.

**Winner: Claude Sonnet 4** for data work specifically.

---

Writing Quality

Professional Emails

**GPT-4o:** Tends to be verbose, occasionally corporate-bland.

  • Example output for "defer client meeting": "I hope this message finds you well. I am writing to respectfully request..."

    • **Claude Sonnet 4:** More natural, better at matching voice. Could tell it my company's tone and it actually matched it.

    **Winner: Claude Sonnet 4** for business writing.

    Creative Content

    **GPT-4o:** More adventurous with structure. Better at breaking the rules intentionally.

    **Claude Sonnet 4:** More consistent tone. Occasionally too cautious.

    **Winner: GPT-4o** for creative, experimental content.

    ---

    Research Synthesis

    Both were given 10 academic abstracts and asked to synthesize themes.

    **Claude Sonnet 4:** Better at identifying contradictions between papers. Produced more structured summaries.

    **GPT-4o:** Faster, but summaries sometimes felt surface-level.

    **Winner: Claude Sonnet 4** for research-intensive tasks.

    ---

    Pricing

    Both ~$5–7/month for our usage level via OpenRouter. Effectively equivalent.

    ---

    Conclusion

    | Task | Preferred Model |

    |------|----------------|

    | Code (speed/completion) | GPT-4o |

    | Code (quality/explanation) | Claude Sonnet 4 |

    | Business writing | Claude Sonnet 4 |

    | Creative content | GPT-4o |

    | Research synthesis | Claude Sonnet 4 |

    | Brainstorming | Either |

    The real answer: use both. They have complementary strengths.

    ← Back to home