Is Claude-3.5 Better Than GPT-4?

Not long ago, I did the same experiment. But back then, the battle was between GPT-4o and GPT-4.

I thought it’d be interesting to use the same prompts I used before.

I’ll be testing GPT-4 against Claude-3.5 on:

• Information Retrieval

• Writing With Contextual Accuracy

• Language Processing

• Creative Storytelling

1/ Information Retrieval

The first task is to summarize an article and provide me with key takeaways. Here’s the prompt I’ll be using with both models.

Summarize article from URL: https://www.anthropic.com/news/claude-3-5-sonnet and provide key takeways.

The first point goes straight to GPT-4. Because well, Claude can’t browse the internet.

Result: GPT-4 winds
Reason: Claude unable to access the web

2/ Writing With Contextual Accuracy

Claude is not off to a good start, but let’s see if the next task changes it.

The next task is to write a persuasive ad while following a set of constraints.

As a direct business copywriter, your task is to write a Facebook ad copy for a [product: "vegan chocolate"] that targets [target audience: "busy moms in their 30s"]. Utilize a [tone: "casual"] and [language: "simple & sarcastic"] that resonate with the audience. At the end of the copy, incorporate a humorous Call-to-Action (CTA) that encourages the audience to take action.

Result: Tie
Reason: Both models had a “good enough“ response while following constraints.

3/ Language Processing

The next task has no real-life use case. It’s about as useful as nipples on a man. But still, it’s a fun way to push the models to their limits.

You'll be given a text. Your task is to replace every 3rd word in that text with the closest synonym. Respond only with a new text.

"One day, Hulk decided he was tired of smashing things and wanted to try something different, so he opened a bakery called "Hulk's Smash Cakes." The cakes were delicious but getting them to the customers in one piece was a challenge since Hulk's gentle touch was still like a minor earthquake."

Result: Tie
Reason: Both models performed the task correctly.

4/ Creative Storytelling

Let’s wrap it up with something creative. In this scenario, I’ll prompt both models to develop a story following my strict details.

Come up with a bedtime story that consists of 10 sentences. 

- The story will have male hero and female antagonist.
- The antagonist will come up with victorious.
- The story will have positive message.
- The story will have humorous ending.
- The story will have simple plot.
- The story will be set in future.
- The story will be written at 3rd grade English level.

Result: Claude-3.5 wins
Reason: GPT-4 not responding with 10 sentences.

5/ Takeaway

I did 4 tests in total. 2 resulted in a tie. Then both models won 1 task each.

But the result (for me) is pretty obvious.

Claude-3.5 is superior to GPT-4 when it comes to writing. The only downside? It can’t access the internet.

Personally, I’m very close to canceling my ChatGPT Pro subscription in favor of Claude. And internet access is the last thing preventing me to do so.

CAN’T MISS

• ChatGPT prompts for brand building

• ChatGPT prompts for fitness

All the prompts are available in our FREE prompt directory.

LET’S HEAR IT!