such wow doge
AI WOW
Published on

Which model should I ask?

Authors

Your "default model" is the one you normally reach for first. Some people call this their "daily driver".1

Your default model should be one of:

  • GPT-5 Thinking (Standard Thinking)
  • Claude 4.1 Opus (Extended Thinking)
  • Gemini 2.5 Pro
  • Grok 4 (Expert) 2

Which of these is best? It's a toss up—just pick the one you like.3

Think of these models as different colleagues, with different strengths and personalities.4 For important tasks, it's often worth asking several models—just like you'd ask several colleagues.

Your default model should always be less than 1 second away.

Model classes: fast, standard, heavy

Let's call the models listed above the "standard" models.

There are also "fast" models and "heavy" models.

Fast models:

  • GPT-5
  • Gemini 2.5 Flash
  • Claude 4.5 Sonnet
  • Grok 4 (Standard)

Heavy models:

  • GPT-5 Pro
  • Grok 4 (Heavy)

Sometimes, a fast model gives an answer that's just as good as a standard model, but... much faster.

Heavy models often take 5-20 minutes to answer. But they're more thorough than the others, and usually more insightful.

As an analogy, think of asking a colleague for:

  • A "quick take" (fast model)
  • A "take" (standard model)
  • A "high-effort, considered position" (heavy model).

Deep Research mode

If you want to exhaustively search the web and get a detailed report, use "Deep Research". Some people strongly prefer Gemini 2.5 Pro for Deep Research. The correct approach is simple: if it's worth Deep Research, it's worth asking all three of Gemini, Claude and ChatGPT.

Appendix 1. The strengths and weaknesses of each model

It's hard to make confident generalisations. Here are some medium-confidence takes:

  • Claude web search is worse than the other models. In addition, Claude just doesn't use its web search tool enough.
  • ChatGPT is the best at web search. 5
  • Gemini 2.5 Pro and GPT-5 Pro are best for Deep Research.
  • Grok is more creative and much less constrained (not afraid of taboo topics, explicit content, etc).

A few more obvious differences:

  • Grok is the only model that's good for searching Twitter.
  • ChatGPT and Gemini have an inline document editor, while Claude does not.
  • Adding Google Docs to Claude is easier than other models—just paste the URL.

Footnotes

  1. Borrowed from car culture, meaning: a car for everyday use. The supercar is for weekends. ↩

  2. Grok is underrated. ↩

  3. Warning: this may change. In spring 2025, ChatGPT o3 was much better than other available models. ↩

  4. Model personality matters. Some people find Claude insufferable, while others love it. By default, models format their outputs differently (e.g. some often use tables and lists, while others prefer continuous prose). ChatGPT and Claude use memory to learn user preferences, and then adapt their personalities accordingly, so people have quite different experiences. ↩

  5. This was obvious a few months ago, and a relatively consensus view. I'm less sure now. Gemini, in particular, has improved. Grok web search also seems good now, but I've not used it much. ↩