I asked Gemini which is better for tax work: ChatGPT, Claude, or Gemini.

1. Handling numbers and processing Excel data (ChatGPT wins by a landslide)
ChatGPT (95 points): The built-in 'Advanced Data Analysis (Code Interpreter)' feature is unrivaled. Because it writes and executes Python code in the background to perform calculations, it has overcome the chronic AI weakness of making mistakes in addition and subtraction. It is the most reliable for tasks such as feeding it tens of thousands of rows of Excel (CSV) bookkeeping data to classify according to specific conditions or to find errors.
Claude (85 points): Its logical data analysis capability is excellent and it creates great visualization tables (Artifacts), but since its environment for executing code directly for computation is not as robust as ChatGPT's, there is a possibility of numerical errors in complex tax calculation formulas.
Gemini (80 points): While its integration with Google Sheets is excellent, it falls quite a bit behind ChatGPT in terms of pure analysis capability to understand and preprocess the structure of uploaded raw data.
2. Interpretation of laws/precedents and report writing (Claude wins by a landslide)
Claude (95 points): When uploading vast and complex Korean tax laws or tax tribunal decision precedents, its ability to reason multidimensionally without losing context is unparalleled. In particular, its ability to write natural and formal reports that sound as if they were written directly by a tax professional, without any machine-translation tone, is overwhelming.
ChatGPT (85 points): While its document writing and summarization capabilities are excellent, it occasionally makes subtle logical leaps in complex legal interpretations or retains a characteristic 'AI translation tone,' requiring practitioners to put in the extra effort to refine the sentences.
Gemini (80 points): It provides a massive context window of up to 1 to 2 million tokens, allowing it to read the equivalent of dozens of books at once. However, it is rated a step below Claude in terms of sophisticated legal causal reasoning and writing fluency.
3. Latest tax law research and integration tasks (Gemini wins by a landslide)
Gemini (95 points): Since it is integrated with the Google search engine in real-time, it is the most accurate for immediately finding annual tax law amendments or National Tax Service press releases released on the same day. Additionally, it integrates natively with existing work tool ecosystems such as Google Docs, Sheets, and Drive, ensuring an uninterrupted workflow.
ChatGPT (90 points): Its own web search function has been significantly strengthened, making its ability to secure the latest information very excellent, but integration with external office tools is limited, so you have to copy and move the results separately.
Claude (70 points): The importance and capability of its web search function are significantly lower compared to the other two models. It is not suitable for researching the latest laws or real-time information and must rely strictly on 'analysis based on uploaded documents.'
Surprisingly objective. lol
‌