Don't use Free ChatGPT (4o) (Update Your Gen AI Awareness #2)
Free Claude, Free Grok, and Free DeepSeek are all better than Free ChatGPT
This is the second in the Becoming an Expert LLM User is not "Prompt Engineering" series of quick articles.
ChatGPT (free) = GPT 4o Isn’t Good
Most people haven’t realized that ChatGPT (with a free account) isn’t a very good model compared to the other options available to them, and that they need to either switch to one of the other free accounts (like Claude, Grok, or DeepSeek) or upgrade to a paid ChatGPT account (which costs $20 per month (i.e. roughly ₹2000 per month in India after you include GST)).
With a free ChatGPT account, you mostly get access to ChatGPT 4o, which is rather bad compared to the other free alternatives. It often misunderstands requests, hallucinates more often, and in general seems to be a bit of a low IQ model.
Claude 3.7 Sonnet Is Excellent
Compared to that, a free Claude account gives you (limited) access to Claude 3.7 Sonnet, one of the best models available (even for people willing to pay the $20 monthly fee that appears to be standard across most LLM companies). 3.7 Sonnet is smart, very sensible, and more grounded than the other models (i.e. it is more likely than other models to say “I can’t answer this” instead of just confidently giving the wrong answer like ChatGPT 4o or even Grok 3 beta).
Claude Can’t Access the Internet (Temporary)
One (temporary) disadvantage of Claude is that unlike the other models, it does not access the internet and hence can’t do a good job of answering some kinds of questions. For example, if your question requires some latest information because of its knowledge cutoff. For example, “Is the Reema Kagti movie Superboys of Malegaon based on a true story?” gets a detailed response from Grok while Claude, whose knowledge cutoff currently is October 2024, complains that “Superboys of Malegaon directed by Reema Kagti isn't actually a film that exists.” Also, if the question is about something that is a bit obscure. For example, for “What is the FutureIQ youtube channel1”, Grok gives a lovely detailed review based on 15 web pages it found from a search, while Claude has nothing—even though FutureIQ has existed for quite some time before October 2024.
It is important to understand why Claude doesn’t know about the FutureIQ youtube channel even though it has existed for 19 months before Claude’s knowledge cutoff. Surely, Claude’s training data would have included some pages about FutureIQ youtube channel. However, an LLM is not a search engine, and unlike a Google search, it does not memorize all the training data. It just “learns” the patterns in the data and only the information about the patterns is stored in its weights. Text like Shakespeare’s sonnets, information about popular movies and celebrities, and other common information appears so many times in the training data that it gets remembered as a pattern. That is why Claude is able to answer the question “What is the Kurzgesagt youtube channel” but not able do the same thing for FutureIQ. By contrast, Grok and even ChatGPT are able to answer it because they use a web search to find information about it and then summarize the results.
But, don’t worry. Free Claude will soon get web search. If you have a paid Claude account and you are based in the US, you should already have web search. Others (including free account users) keep checking your Claude profile settings page, and if “Web Search BETA” has appeared there, enable it.
Grok is Good
By now you should also have realized that Grok is another LLM you can use which is better than free ChatGPT—either because you were paying attention in the earlier paragraphs or hopefully because you read my previous post in this series. (If not, why not? Please subscribe!)
I should point out that if you can afford it, ChatGPT Plus (the paid version) is very good value for money. It gives you access to ChatGPT 4.5 which is probably a better model than all the others available at this price or free (except that Claude 3.7 Sonnet might be better for programming and Grok might be better for finding latest/controversial information), and you get 10 “Deep Research” reports (mind-blowingly good, but I’ll cover this in a future post), and o3-mini-high and o1 both excellent models but I will not talk about them right now because I don’t want to overload your brain.
So anyway, either upgrade to ChatGPT Plus or switch to Claude or Grok.
Not-so-subtle self-promotion. Please check it out.
Amazing, I agree there's nothing like 3.7 sonnet,
But then why gpt4o is among top 5 in LMSYS leaderboard
Wow, very informative and useful
I have a perplexity paid plan, but I feel when I use Claude through perplexity and through my free claude account on claude website, crux and tone of the outputs are different most of the times
I do not know how to make claude or any other model in perplexity show output similar to the one it shows in actual website of their respective models
I will be grateful if you write a blog covering this question.