ChatGPT vs. Claude: The Ultimate Comparison

ChatGPT and Claude are currently two of the most exciting and best AI chatbots on the market. But which one is better?

To find out, I extensively tested and compared both tools, including usability, feature set, image recognition, model quality, and programming.

The result:

It depends on what you want to do with it!

If you value easy operation, many features, and an appealing user interface, ChatGPT is clearly the better choice.

For more complex tasks or longer texts, Claude might be the better choice. I find the new language model "Claude 3 Opus" better than "GPT-4 Turbo" for many tasks.

But let's look at this in detail:

Comparison Table

1. Usability: ChatGPT Wins

When it comes to usability, ChatGPT is several steps ahead of Claude.

For example, with ChatGPT it's possible to edit prompts after the fact (by clicking the small pencil under the prompt):

I also find it unfortunate that you can't share chat histories with others in Claude, which is easily possible with ChatGPT through the share function:

2. Features: ChatGPT Wins

What applies to usability also applies to the feature set:

Here too, ChatGPT clearly has the edge. The following features are not available in Claude:

• A web browser (although it must be said that ChatGPT's web browser isn't particularly good)
• A code interpreter
• Image generation
• The ability to create custom instructions
• GPTs

It also offers better settings options. For example, you can set the interface language, two-factor authentication, archive chats, or delete all chats at once:

For businesses, ChatGPT additionally offers Team and Enterprise plans that provide higher limits, team features, and better security and privacy.

Claude only offers a free plan and the Pro plan.

3. Image Recognition: ChatGPT Wins Narrowly

Recognition of text and elements in images works somewhat (but not much) better with ChatGPT than with Claude 3 Opus.

Here's an example of extracted handwritten notes with ChatGPT:

With Claude, the result was okay but not quite as good. Claude especially didn't handle the first two points well. Additionally, Claude doesn't correct obvious spelling errors:

4. Model Quality: Claude Wins

After extensive testing, I can say:

Claude 3 Opus beats GPT-4 Turbo in most cases. But let's look at this more closely:

4.1 Context Window & Understanding

With Claude 3 Opus, it's possible to insert an incredible 200,000 tokens as a prompt and provide it as context to the AI model. That's about 800,000 characters or approximately 150,000 English words.

With GPT-4 Turbo, only a meager 32,000 tokens (about 128,000 characters or 24,000 English words) are possible. That's about one-seventh of Claude.

And this opens up many new use cases. For example, you can ask Claude questions about very long documents, which it can answer much better than ChatGPT, which uses RAG (Retrieval Augmented Generation) for documents:

Unlike ChatGPT, Claude also doesn't tend to forget the answers at the beginning of a chat as the chat gets longer.

But that's not all:

Claude is also better at processing the information it receives in the prompt. That means "understanding" it as a whole or finding individual pieces of information in a huge amount of text (even when asking ChatGPT within its context window of 32K tokens, i.e., without using RAG).

4.2 "Laziness"

GPT-4 Turbo has a big problem that OpenAI hasn't been able to fix since its release in November:

It's sometimes lazy, stubborn, or both. And this manifests as follows:

• It has trouble following very complex instructions
• It tends to write significantly shorter texts than predecessor models (i.e., GPT-4 without Turbo)
• When coding, it often uses placeholders instead of showing you the complete code (even after multiple requests)
• It sometimes completely refuses to work, especially when you've already generated a lot of text in a chat

Here's an example chat where I wanted to write a longer blog article section by section. Already at the 4th chapter, ChatGPT refused to work:

This makes it extremely annoying to work with for certain tasks. Especially if you want to use it for programming, creating complex GPTs, or writing blog articles (like me).

Claude doesn't have any of these problems.

4.3 Repetitions and Writing Style

All OpenAI AI models, including not only GPT-3.5 but also GPT-4 and GPT-4 Turbo, tend to repeat themselves.

Here's an example text from one of my clients where every single sentence (yes, every one!) begins with "You can":

Daisy Salad: You can make a delicious salad from daisy leaves and flowers. You can also add other wild herbs, such as dandelion, nettle, chickweed, or sorrel. You can dress the salad with olive oil, lemon juice, salt, pepper, and honey. You can eat the salad as an appetizer or as a side dish.

Daisy Smoothie: You can make a refreshing smoothie from daisy flowers and leaves. You can also add other fruits, such as bananas, apples, pears, or berries. You can thin the smoothie with water, milk, yogurt, or juice. You can drink the smoothie for breakfast or as a snack.

Daisy Soup: You can make a warming soup from daisy leaves and flowers. You can also add other vegetables, such as potatoes, carrots, onions, or celery.

With Claude 3 Opus, repetitions also occur. Like most other LLMs, it sometimes creates unnecessary summaries at the end of texts or text sections.

However, it repeats itself significantly less than GPT-4 Turbo or GPT-4.

4.4 Chatbot Arena Leaderboard Placement

Claude 3 Opus was the first model to beat GPT-4 in the Chatbot Arena Leaderboard (at least until the release of gpt-4-turbo-2024-04-09 on April 9, 2024):

Even Claude 3 Sonnet, the smaller sibling, was able to beat two older GPT-4 models and GPT-3.5.

5. Coding: Claude Wins

I've now written half a dozen Python scripts, two WordPress plugins, and lots of JavaScript with Claude, and I'm impressed.

Claude has made very few errors when coding. While I often had to go through many debugging loops with GPT-4 until a script or program runs smoothly, I accomplish it with fewer attempts with Claude.

It also rarely omits code parts or completely refuses to work, which GPT-4 Turbo in particular likes to do.

When coding, it's also a huge advantage that Claude has a super long context window and you can include complete documentation in the prompt (which it can then incorporate very well).

6. Accessibility & Free Version: ChatGPT Wins

Both Claude and ChatGPT offer free plans, which is great. With Claude, "Claude 3 Sonnet" is available as the AI model, and with ChatGPT, "GPT-3.5".

Since Claude 3 Sonnet is somewhat better than GPT-3.5, Claude wins in the free version category.

A major drawback of Claude unfortunately is:

The Pro version is not officially available in all countries. You can still subscribe by using a VPN and entering a US address, but this puts it at a disadvantage compared to ChatGPT:

ChatGPT is available in all versions (Free, Plus, Team, and Enterprise) worldwide. The user interface can also be set to your preferred language.

Since April 1, 2024, ChatGPT can also be used without logging in.

My Conclusion

ChatGPT is the better tool than Claude. But Claude 3 Opus is, in my opinion, better than GPT-4 (and Claude 3 Sonnet is better than GPT-3.5).

And no:

This doesn't mean I have stopped or will stop using ChatGPT. I still use it daily.

I simply prefer using Claude for certain tasks. These include writing longer texts (especially blog articles), complex tasks with long prompts, or coding.

And I think this trend will continue in the future: there won't be one chatbot that does everything best, but rather you'll achieve the best results by combining different chatbots.