What ChatGPT model to use

At the moment, there are seven(!) different models of ChatGPT only. But what model to choose?

Besides OpenAI numerous models, there are also several models per Claude from Anthropic, Google Gemini, xAI Grok. Are the new models better than the previous ones? Not exactly. They are all slightly different and the choice depends on the task at hand. Andrej Karpathy, co-founder and ex-OpenAI, made a nice summary of his use cases, and I thought I will provide my own observations, as well as comparison of different models.

90% of all tasks can be completed by GPT-4o

This is OpenAI’s default, general-purpose model, where “o” = omni, meaning it is multimodal and can process images and audio in addition to text. It is good to start with for any kind of task, from recipes to product roadmap, and it has the best balance of speed, capability, and flexibility. If you are not satisfied with the answer or need more insights, then proceed to the next best tool->

o3 covers the rest of 10% needs and its share is increasing

I found myself to be increasingly often using the o3 model, which is a reasoning model, as it takes time to think and question its own thinking process. It is very effective for tasks that require factual accuracy and is suitable for deep, multi-step analysis, coding, math, science, or strategic planning. It has access to all ChatGPT tools, like browsing, Python, file uploads, vision, image generation, but the trade-off is that it’s slower, and can be thinking for 5-10 minutes depending on the task.

Deep Research mode gets you a quality report on any topic in minutes

It’s not a model, but toggle you can select in any model, which triggers ChatGPT autonomous agent to spend 5-30 min browsing hundreds of sources, and then deliver a fully cited report. Personally, I’ve used it to research topics of interest or for learning new concepts/areas. You can also set the tone for the report depending on your audience or preference. Also, the quality of your report will depend on the research questions, so take your time to formulate questions you care about, which will guide the agent’s work.

I also use “Temporary” chat option to avoid saving every minor conversation, making it easier to find old useful chats.

Other ChatGPT models worth mentioning

GPT-4.1
It’s like GPT-4o’s speed combined with o3, offering stronger logic handling and the ability to process up to one million tokens. This is suitable for entire books, large PDFs, or lengthy conversations. Best used for programming, technical writing, and precise instruction-following. It requires less prompt tuning and is built for reliability over long contexts. Difference with o3 is the speed.

GPT-4.5
OpenAI claims it’s good at “creative writing, nuanced advice, and high-stakes content like reports and strategy documents”, but I haven’t figured out how to get results from it. Maybe it can be good when ideating/editing whole chapters or merging books, etc.

OpenAI o4-mini, 04-mini-high, 4.1- mini, 4.5 - useless

Use this quick reference to match your task with the right model:

• Everyday questions, need images, audio, browsing, or file tools? → GPT-4o.

• Need fact-checked and high-quality thinking? → o3.

• Working with code or >32 k tokens of text? → GPT-4.1 (API).

• Need fresh, sourced info and 20+ pages ? → Deep Research mode.
  

AI Competitors at a Glance

Other major AI platforms also offer unique advantages, but their selection is rather a matter of taste. Here’s how they compare:

Claude (Anthropic)
Having subscriptions of both ChatGPT and Claude is kind of useless, so I stopped using Claude, but I liked Claude's tone and technical answers. It offers 100 k+ context, rigorous logic, tidy structured answers. Use it when summarising whole books or long codebases, as well as step-by-step reasoning tasks.

Google Gemini
Lately, very capable models on par with ChatGPT (2.5 Flash, Pro), has access to all Google tools and is integrated through the apps. A bit wordy for my taste, and is not as “smart” for daily work tasks, but getting better and better.

xAI Grok
Very fast and direct, especially for real-time information. Best for quick trend checks, live statistics, and alternative perspectives.

Perplexity AI
Feels like a hybrid between Google and Wikipedia. Use it for research, reports, or fact-checking ChatGPT’s answers. Highly factual and accurate, but less creative.

TL;DR

Use GPT-4o as your default model. Switch to GPT-o3 for advanced logic and analysis. Choose GPT-4.1 for coding or long documents. Use GPT-4.5 for creative brainstorming (double-check it). Activate Deep Research mode for tasks that require up-to-date information and web research. Use Temporary chats to keep the workspace tidy.

---

I created a useful PDF 100 Ready-to-Use GPT-4o Prompt Workflows for Busy Professionals, designed to save you time and make ChatGPT actually useful in your day-to-day work. You can download it for free here. By downloading, you'll also subscribe to my newsletter with weekly posts (if not already subsribed). You can unsubscribe anytime.