Claude vs. GPT: Which is best for note-taking?

Note-taking workflows in Reflect Notes can switch between Anthropic’s Claude 3.5 Sonnet and OpenAI’s GPT-4o, and the practical takeaway from side-by-side tests is that Claude 3.5 Sonnet tends to produce better-quality text—especially when the task demands nuance, counterarguments, or clearer rewriting—while GPT-4o more consistently improves formatting and sometimes compresses content more aggressively.

The comparison starts with Reflect Notes’ model controls: users can leave the AI provider on the default (Anthropic), or manually set it to Anthropic (Claude 3.5 Sonnet) or OpenAI (GPT-4o). Reflect Notes also updates its default model over time, so the “best” choice can shift as providers roll out new versions. Still, when forced to pick one model for note-taking tasks, the guidance is to stick with Claude 3.5 Sonnet.

In a short summarization test, both models deliver usable summaries, but Claude 3.5 Sonnet retains more information while GPT-4o condenses more. The narrator credits Claude for being the more informative summary, even if it could stand to compress further.

When asked to generate writing—specifically drafting a corporate-style email—the results diverge. Claude’s output includes more “fluff” and a more generic opening, which can slow down real-world use if the goal is a clean, ready-to-send message. GPT-4o, by contrast, produces a cleaner email with better formatting and more directly usable structure, even though it may be shorter.

The biggest quality gap appears in logic-focused work. Given a claim that AI will fully automate 30% of jobs by 2030, Claude responds with a structured set of counterarguments, framing the prediction as likely overstated and listing concrete reasons. GPT-4o answers in a more paragraph-like form and leans toward softer phrasing. For tasks that require argumentation and persuasive clarity, Claude’s list-style counterpoints come across as more compelling.

For “simplify and condense” tasks, GPT-4o again shows strength in presentation: it turns complex material into a step-by-step process with clear formatting. Claude also simplifies well, but GPT-4o’s reformatting makes the content easier to follow. In the CNC manufacturing example, both models help a non-expert understand the topic, yet Claude is still favored overall because it produces a more useful distilled explanation.

Finally, in a rephrasing test using the narrator’s own paragraph, the models land close. GPT-4o’s version is judged worse, while Claude’s is judged better, though the critique notes that the evaluation is limited because the prompt was a simple rephrase rather than a targeted rewrite.

Overall, the recommendation is to keep Reflect Notes on the default model (Claude 3.5 Sonnet) for note-taking, while recognizing that GPT-4o can be preferable when formatting and step-by-step structure matter most. A follow-up comparison is promised for “chatting with your notes,” where the model choice may affect retrieval and conversational behavior differently.