GPT-5 vs Claude 4 vs Gemini 2: Real-World Work Test Comparison
Results from testing the three models on the same tasks (data analysis, code review, document summarization) over the past three months.
1. Data Analysis Ability
- GPT-5: Excels at creating tables and statistical summaries. Also handles R coding adeptly.
- Claude 4: Strong at reading long CSV files and deriving insights. However, weak at catching errors in Korean code.
- Gemini 2: Integrates well with Google Docs, making it convenient to use with collaboration tools. Quite fast for simple queries.
2. Code Review Accuracy
- GPT-5: Frequently catches security vulnerabilities. Downside: too many comments.
- Claude 4: Specialized in finding logic errors. Detailed explanations are a plus.
- Gemini 2: Fastest speed but somewhat lower accuracy.
3. Document Summary Quality
Personally, Claude 4 felt the most natural. GPT-5 was too rigid, and Gemini sometimes missed the key points.
Conclusion: It seems the answer is to use all models interchangeably depending on the situation. Each has its own clear strengths.
7 answers
I feel the same way
But what about the cost?
For data analysis, GPT is good, but for Korean document summarization, Claude is really natural. That's why I use them interchangeably. However, Gemini is only useful within the Google ecosystem.
Well, I actually thought GPT-5's summaries were fine? Claude was a bit inconvenient when catching Korean errors. But overall, I do agree with some parts.
Oh, thanks for the great info.
I've tried it too, and Claude is definitely better for code reviews. GPT talks too much lol.
Do you have a source? I'd like to try it too.