Which AI model shall we chose?

Published

May 22, 2025

In what follows, here is my personal take as of date:2025-05-22

Generative AI based on Large Language Models (genAI) is great for many tasks. In this course we only focus on aspects of Data Analysis:

designing analysis
writing code
data wrangling such as joining tables, sample design, variable transformations
exploratory data analysis
modelling, regressions, machine learning
causal inference
creating tables and graphs
writing reports and presentations

Different AI providers and their models

OpenAI ChatGPT

The main current models are 4o, o3, and 4.1

Here’s an updated, compact guide including o3, o4-mini-high, GPT-4.5, and Deep Research:

Here ChatGPT’s quick guide to when and why to use which model

🔍 Model Overview

Model	Strengths	Use When
GPT-4-turbo (4o)	Fast, accurate, handles long prompts and code well	Default for coding, EDA, modeling, teaching
GPT-4.5	Slightly better reasoning and math; not always faster	More complex logic, multi-step planning
GPT-4 (base)	Stable, reliable for structured work	You need consistent responses (e.g., templates)
o3	Compact, efficient, more creative but can be fuzzy	Brainstorming, creative prompt design
o4-mini-high	Lightweight, fast, good for quick checks or when resources are limited	Instant feedback, code sketching
Deep Research	Access to full documents, citations, deep factual retrieval	Literature reviews, technical deep-dives

✅ Best Model by Task

Task	Best Model(s)	Notes
Designing analysis	4o / GPT-4.5	Handles multi-step reasoning well
Writing code (R, Python, SQL)	4o / GPT-4.5 / o4-mini-high	Use 4o for tidyverse-heavy tasks; o4-mini for quick draft
Data wrangling	4o	Very good with `dplyr`, `data.table`, `pandas`
Exploratory data analysis	4o + code interpreter	Visuals, summaries, and diagnostics
Modeling (ML, regressions)	GPT-4.5 / 4o	Clear, structured models and diagnostics
Causal inference	GPT-4.5 / 4o	Handles DiD, IV, RDD, matching logic well
Creating tables and graphs	4o / GPT-4.5	Knows LaTeX, Markdown, ggplot2, matplotlib formatting
Writing reports / slides	4o / GPT-4 / o3	4o for clarity, o3 for more creative text generation
Literature search / citations	Deep Research	Finds, summarizes, and cites academic papers

Anthropic Claude

The main current model is Claude Sonnet 3.7.

Key tools * Projects: organize files, allow inquiry. One example is full codebase.

Others

There are many other models, but I have much less experience.

Free vs Pro?

The current free models are great for many tasks such as coding, idea generation.

ChatGPT

The free version offers: access to GPT‑4.1 mini, real-time data from the web with search. Plus * Limited access to GPT‑4o, OpenAI o4-mini, and deep research * Limited access to file uploads, data analysis,

The Plus version offers

access to reasoning models (OpenAI o3, OpenAI o4-mini, and OpenAI o4-mini-high)
access Deep Research
higher limits on advanced features: file uploads, and data analysis
access to a research preview of GPT‑4.5
access to GPT‑4.1, a model optimized for coding tasks
can create and use projects, tasks,

Claude

The free model can be used for chat and data analysis.

The paid tier for Claude

More usage – for details see limits
access to Projects to organize chats and documents
web access
extended thinking for complex work

Other cool stuff I use

Notebook LM

[Google’s Notebook LM] (https://notebooklm.google/) is able to “understand” and summarize any material (such as research paper) and relate it to other topics. Can create fun audo summaries like a podcast. Here is one on a research paper of mine on cultural homophily.

Github Copilot

Github Copilot goes inside your code editor such as Rstudio, VSCode, Jupyter Notebook and helps writing code. Great to write frequent stuff like loops or graphs.

It has an Education – free access for students: GithubEducation

Cursor

Cursor AI is the most popular AI code editor, I have very limited experience, but is favored by software developers.

Feedback

Dear Reader. I have limited experience. Suggestions are welcome, please post an issue.

--- title: "Which AI model shall we chose?" date: "2025-05-22" --- In what follows, here is my personal take as of `date:2025-05-22` Generative AI based on Large Language Models (genAI) is great for many tasks. In this course we only focus on aspects of Data Analysis: * designing analysis * writing code * data wrangling such as joining tables, sample design, variable transformations * exploratory data analysis * modelling, regressions, machine learning * causal inference * creating tables and graphs * writing reports and presentations ## Different AI providers and their models ### OpenAI ChatGPT The main current models are `4o`, `o3`, and `4.1` Here's an updated, compact guide including **o3**, **o4-mini-high**, **GPT-4.5**, and **Deep Research**: *Here ChatGPT's quick guide to when and why to use which model* #### 🔍 **Model Overview** | **Model** | **Strengths** | **Use When** | | -------------------- | ---------------------------------------------------------------------- | ----------------------------------------------- | | **GPT-4-turbo (4o)** | Fast, accurate, handles long prompts and code well | Default for coding, EDA, modeling, teaching | | **GPT-4.5** | Slightly better reasoning and math; not always faster | More complex logic, multi-step planning | | **GPT-4 (base)** | Stable, reliable for structured work | You need consistent responses (e.g., templates) | | **o3** | Compact, efficient, more creative but can be fuzzy | Brainstorming, creative prompt design | | **o4-mini-high** | Lightweight, fast, good for quick checks or when resources are limited | Instant feedback, code sketching | | **Deep Research** | Access to full documents, citations, deep factual retrieval | Literature reviews, technical deep-dives | ### ✅ **Best Model by Task** | **Task** | **Best Model(s)** | **Notes** | | ----------------------------- | --------------------------- | --------------------------------------------------------- | | Designing analysis | 4o / GPT-4.5 | Handles multi-step reasoning well | | Writing code (R, Python, SQL) | 4o / GPT-4.5 / o4-mini-high | Use 4o for tidyverse-heavy tasks; o4-mini for quick draft | | Data wrangling | 4o | Very good with `dplyr`, `data.table`, `pandas` | | Exploratory data analysis | 4o + code interpreter | Visuals, summaries, and diagnostics | | Modeling (ML, regressions) | GPT-4.5 / 4o | Clear, structured models and diagnostics | | Causal inference | GPT-4.5 / 4o | Handles DiD, IV, RDD, matching logic well | | Creating tables and graphs | 4o / GPT-4.5 | Knows LaTeX, Markdown, ggplot2, matplotlib formatting | | Writing reports / slides | 4o / GPT-4 / o3 | 4o for clarity, o3 for more creative text generation | | Literature search / citations | **Deep Research** | Finds, summarizes, and cites academic papers | ### Anthropic Claude The main current model is Claude Sonnet 3.7. Key tools * Projects: organize files, allow inquiry. One example is full codebase. ### Others There are many other models, but I have much less experience. ## Free vs Pro? The current free models are great for many tasks such as coding, idea generation. ### ChatGPT The free version offers: access to GPT‑4.1 mini, real-time data from the web with search. Plus * Limited access to GPT‑4o, OpenAI o4-mini, and deep research * Limited access to file uploads, data analysis, The [Plus version](https://openai.com/chatgpt/pricing/) offers * access to reasoning models (OpenAI o3, OpenAI o4-mini, and OpenAI o4-mini-high) * access Deep Research * higher limits on advanced features: file uploads, and data analysis * access to a research preview of GPT‑4.5 * access to GPT‑4.1, a model optimized for coding tasks * can create and use projects, tasks, ### Claude The free model can be used for chat and data analysis. The paid tier for Claude * More usage -- for details see [limits](https://support.anthropic.com/en/articles/9797557-usage-limit-best-practices) * access to Projects to organize chats and documents * web access * extended thinking for complex work ## Other cool stuff I use ### Notebook LM [Google's Notebook LM] (https://notebooklm.google/) is able to "understand" and summarize any material (such as research paper) and relate it to other topics. Can create fun audo summaries like a podcast. Here is one on a [research paper of mine on cultural homophily](https://notebooklm.google.com/notebook/655d48f9-bff7-4ac3-908b-5b9bc48e5be4/audio). ### Github Copilot Github Copilot goes inside your code editor such as Rstudio, VSCode, Jupyter Notebook and helps writing code. Great to write frequent stuff like loops or graphs. It has an Education -- free access for students: [GithubEducation](https://github.com/education/students) ### Cursor [Cursor AI](https://www.cursor.com/) is the most popular AI code editor, I have very limited experience, but is favored by software developers. ## Feedback Dear Reader. I have limited experience. Suggestions are welcome, please post an issue.