Claude vs GPT

I have been using Claude and GPT in my knowledge work. Here is how I split the work between the two models to leverage their comparative advantages:

1. Claude for deep research, mainly for resource gathering. GPT can supplement, but for a clear and well-architected view of a field, Claude is much better.
2. GPT for brainstorming and in-depth discussion. GPT will typically come up with more innovative ideas. Moreover, GPT is very good at summarizing its responses or insights in a single, concise sentence.
3. GPT for summary. GPT is much better than Claude at summarizing papers. GPT will cover necessary background knowledge even when it is not explicitly a section of the paper, and thus the summary itself is more understandable. My theory is GPT is better at representing the logic of the paper towards understanding.

One small thing that I think both models are weak at: they can’t actively illustrate. A concept that can be explained well with a good conceptual illustration is often hard to convey even with 200 words. Both models just use text and don’t produce illustrations on their own. Even when asked to, the quality is not good enough.