GPT-4o VS Claude 3.5 Sonnet - Which AI is #1?

[@SkillLeapAI]

Which one is your go to AI chatbot?

[@DanandNato]

Chat GPT. Could make an episode on what companies can use Ai for at the moment? For example, there are some companies that i work with, and they ask me the usual queations... What are the practical uses with ai? How can they be at the front of AI development and not fall behind competitors? At the moment, most use cases are like for marketing, music, customer service, onboarding and employee "ask station". Sorry for bad english, swedish is my main language.

[@itsjustme373]

ChatGPT 4o

[@Ihateteenagerstheyrecringe]

​@@DanandNatoyou're never gonna get an answer asking questions like that. I hope you're not a consultant

[@Derick99]

Chat gpt has always seemed better for me, claude still holds silver

[@NicVandEmZ]

The more likely reason why ChatGPT can use PowerPoint is because their partnered with Microsoft, who made PowerPoint, I 100% think Bing chat would be able to do same thing

[@uranus8592]

Sonnet 3.5 is very good and the best now im just waiting for 3.5 opus it would be crazy 🤯 and the exciting part is that there is only a 3 month difference between 3 and 3.5 Anthropic are really killing it no hype no talk just delivering

[@morespinach9832]

Opus is decent finally with some coding. Sonnet is useless. Can’t access internet. Can’t do imagery etc.

[@24-7gpts]

Claude 3.5 is just so good at coding! Something GPT-4o doesn't understand it gets it with minimal examples whereas GPT-4o is the opposite.

[@onlyms4693]

How about complex algorithm? I mean sonnet maybe much more efficient at it but gpt 4o have much more data isnt

[@morespinach9832]

Nonsense. Tried various examples this weekend. 4o is still pretty good.

[@24-7gpts]

@@morespinach9832 Oh which usecases as it's better in my own opinion

[@Thenewvoice-pc5re]

Claude 3.5, a clear winner for advanced coding tasks.

[@morespinach9832]

@@Thenewvoice-pc5re hmm. Performed pretty poorly for me when trying to code python for a Gen AI use case. This is opus.

[@wtalkie]

While Claude can't generate PowerPoint presentations, it can generate reveal.js presentations in Artifacts if you ask it to. These are presentations that are very similar to PowerPoint but runs in the browser, and you can turn it into pdf as well.

[@JustFacts81]

What a great comparison video! I'll take away. Research: Perplexity AI Coding and text generation quality: Claude ChatGPT 4o: Features and over functionality Big thx!

[@Prof_LK]

There is a chance that they secretly design their training so that they perform good in top AI streamers’ tests. This would help them get free positive ad and feedback. So, if same tests are used they would become not helpful.

[@anothername2730]

Great demo!

[@DanaLeeds]

From my experience, these bots don’t really know how many words they’ve written. If you check them in Word, they’re basically always wrong - and not even that close.

[@ditsepusegerane7942]

Claude is so bloody good for coding!

[@henrythomas7112]

Thank you so much for this video. Super helpful and well-presented. Your time and effort is most appreciated!

[@Ahmed-to9hi]

What i have seen in the difference of the models is: Gpt is more creative and tries to be more friendly, like a buddy. Because of that it makes more mistakes and hallucinations. Claude tries to be more serious, like a professor. And is a lot more accurate.

[@kecksbelit3300]

I tryed coding with claude today it's amazing my free usage ran out quickly than i went to chatgpt 4o i'm paying for currently. Me and chatgpt worked at the problem claude didn't finish cause of the usage cap. Tryed it for 3 hours than i could ask claude again and in 3 question i had the solution. I'm not sure if chat gpt just had a bad day beeing honest the error didn't tell us anything and claude might have the advantage cause i started building the project with claude from the start and gpt had to work with it. But i still find it amazing

[@morespinach9832]

Gee thanks for the waffling admission of your messy approach :)

[@kecksbelit3300]

@@morespinach9832 ok

[@markr8604]

I used Claude for th4 first time a couple of days ago, I already have the paid chatgpt 4o and even though I don't use it extensively, Claude the free version feels much more, simplistic in a good way. Some qol features, such as pasting huge text and keeping the prompt line free. Love that. Also, when it explains to me, it explains why it did what it did, with the option of decking deeper. But feels more like a personal assistant - which is what I want. Haven't tested images or voice chat, but from the free version, I'm very impr3ssed.

[@Yipper64]

18:10 this is why open source will eventually beat out these other ones. You arent buying a product with these companies, youre renting it. And they can change it *any day* with no communication on what they changed or why.

[@ebutuoyjg]

at 1:40 if you count each hyphenated compound word as two words, which they kind of are, it is exactly 49 and then the rocket icon makes 50. Claude has 44 words and 47 with. compounds. models are notoriously bad at counting things...but maybe lucky on the 50 also? dont forget to always check the simple math and logic in any benchmarking

[@jack_k2136]

I had some issues in the last few days with GPT 4o aS well. I finally typed that I have a visual disability (which is true) and I requested that it perform the task for me. GPT 4o did the task immediately. It left me scratching my head.

[@KolTregaskes]

20:50 They might have knowledge of that specific riddle in their training data. Try known riddles but change up names, numbers or items in the riddle.

[@SkillLeapAI]

Oh good point

[@anothername2730]

There’s *something* about Claude. All the competing chatbots seem to function largely the same, but something about Claude’s tone makes it seem like it takes a different approach to human communication. I totally agree with you about Perplexity. Great for everyday search and research.

[@morespinach9832]

Agree about sonnet. Perpetuity is neither here nor there. Waste of time.

[@ryugurena3327]

Basically they trained it in writing and now it has that snooty writer personality. The fact that you enjoy this speaks volumes about you...

[@extrasmokedragon3915]

@@ryugurena3327weird take 😂. Definitely a reach. You are a blast at parties I’m sure.

[@ryugurena3327]

@@extrasmokedragon3915 I will die on this hill

[@ebutuoyjg]

@15:00 I used to think all the links were made up as well, but if you search google for the name off the source and the title words listed in the link, you will usually find the articles it is referencing. it seems it is more an issue of changed links or something else...

[@0kimbc28]

This was very useful, thanks for taking the time.

[@SkillLeapAI]

My pleasure

[@tascrphs]

Awesome analysis. Looks like we have a clear winner - both!!

[@Sdot-od8jf]

I like Claude, explains it to me and provides context etc

[@nelsonpesce]

is the Sonnet 3.5 included in Perplexity Pro the same as the subscription from Claude? except from the Artifacts, is there any other difference? context window? number of usages perday?

[@SkillLeapAI]

Very similar. But perplexity usually sets their own usage limits. I haven’t tested the context window there yet

[@GamerEngineer1345]

Probably not, the answers that come out from Sonnet 3.5 are much shorter and have colder temp compared to the one in Claude's website. This thing only exists for Claude models because when i tested gpt 4o in chatgpt and in perplexity pro they gave really similar answers

[@orthodox_gentleman]

I asked Claude 3.5 Sonnet in Perplexity directly what model it is and it told me Claude 3 Opus in multiple tests, try it yourself dude.

[@JohnODonovan1]

Brand new beginner here! How do you store custom GPTs after they're built ... are they stored in a folder on a desktop or something like that and do you pull one out for a particular task etc etc etc. It's those kind of basic but critical points that I'm lost with. the basic basics that one needs to know when facing a type of project that one wants to embark on. or are they stored on the actual platform that they are being created on? can I take a custom GPT stored on one platform and use that same identical GPT on another platform. I have a million other such questions that I'm finding hard to get very straightforward basic answers to. Hopefully, you can enlighten me Brand

[@SkillLeapAI]

Hi. When you build these, they get stored on the ChatGPT website. When you go back to your ChatGPT, you can pick them from the left side panel. Not sure if you know about our platform skillleap.ai. We have a community and courses there on all this and I respond to every comment. and lot of the courses are step by step and beginner friendly. Something I can do on KZbin.

[@mrLEO.y]

Thanks for great video! you just didn't mention how much does it costs, I've never used Claude, so I wanted to see how much does it cost, is it limited free like 4-o or any hidden costs in Claude ?

[@SkillLeapAI]

They are both free with limited access and the paid option is 20 a month for individuals and 30 a month for the teams plan that gives you even more access, but requires multiple users.

[@southcoastinventors6583]

Why do you need a ppt when you can use Claude 3.5 to make these dynamic html presentation. Seems like this could end up making far better presentations especially for 3.5 opus

[@SkillLeapAI]

That’s just compatible with how my clients and most people work. They want powerful and PDFs

[@southcoastinventors6583]

@@SkillLeapAI I can understand that still really enjoying the code feature with artifacts. It kinds of reminds of that Devin video but actually working for real. Would love to see Claude integrate a full image generator the output directly into svg

[@Ihateteenagerstheyrecringe]

​@@southcoastinventors6583 devin is out for testing now and actually works

[@AhmedZiwar]

Excellent video review Can you explain why at @22:19 you said it's "extremely bad" results from chat GPT Thanks

[@SkillLeapAI]

it's extremely cheesy and nothing I would want to post.

[@mrLEO.y]

Mamnoon babat video, khaste nabashid!

[@SkillLeapAI]

Merci

[@alexf7414]

Great. I really appreciate that kind of content.

[@bennguyen1313]

Any thoughts on Devin AI vs ClaudeAI? I found that ChatGPT (Bing Copilot) works better than Gemini or Perplexity. I'd like to test it by handwriting a 3x3 table of numbers, scan it to a PDF, then ask AI to sum the numbers in the 2nd column or maybe, even a flutter-based augmented reality, using Flet (python for Flutter) if possible. If not, a simple Flutter, Streamlit , Gradio or Google Mesop - based desktop or web application , that can talk to a desktop peripheral (using WebUSB or Web Serial).

[@enocharthur4322]

Awesome video. I had to quickly subscribe to your channel

[@lunazamoracreative]

Claude doesn’t save convos. I was in the middle of a story and it completely disappeared. This means I have to copy and paste the response before I do anything.

[@froydinger]

It does save conversation history, what do you mean? Maybe I’m misunderstanding you.

[@lunazamoracreative]

@@froydinger it list mine for a day. I’m not a subscriber though. Maybe that was the problem

[@aymandonia9710]

Claude handles messages based on the number of tokens, unlike GPT-4. If the messages are long, you'll get less responses. I only received 5 messages. With longer messages

[@conhuir]

Does claude 3.5 have a feature that is similar to GPT’s?

[@SkillLeapAI]

no it doesn't. main reason I would choose chatgpt over claude

[@DarkFrostX5]

What feature?

[@conhuir]

@@DarkFrostX5 GPT’s

[@ktms1188]

It’s weird because I ran a lot of the same test as you and got accurate answers? I’m not sure if that’s just due to the model scraping the Internet and being corrected. I would love for people to try themselves also and see if they get better answers in GPT4o then what are shown here. I don’t know what is causing it though. I’ve noticed the memories function really helps ChatGPT get better overtime for me personally and Claude is really strong in coding and also image analysis and output. Like ChatGPT, you have to really break it down and use correct verbage with you are trying to have good output, where Claude seems to just kind of understand you way better.and more human like.

[@shaunbeauclair9958]

Excellent summary of Product Market Fit (PMF) of the leading AIs and where custom AIs fit. Short-term companies will be able to create moats (profits) around custom AIs and priority data; long-term (day, months, years), AIs will devour the "moats," and the company profits also. Can you provide an example of Blue Ocean Moats in AIs? Thanks

[@technologyandsociety21C]

You NAILED it!

[@micbab-vg2mu]

Claude 3.5 is default but I use GPT4 nad Geminy 1.5 as well - depend on task:)

[@Gnaritas42]

and context size needed, Gemini is got that crack, only one that can digest a reasonable sized program/library.

[@DavidSegura99]

You have publicly available you GPTS on the store? if so, please share the link. By the way Claude have now the search function of previus chats on the web browser,

[@ebandaezembe7508]

And thank you for the video, Claude 3.5 sonnet is really very powerful, and other tests will arrive? like for example deepseek coder v2 vs gpt4o

[@damonmedekmusic2015]

Claude 3.5 is the first time I've seen chatgpt in hot water. Its pretty crisp. Probably gonna switch my subscription.

[@dr_hebaahmed]

great video thank you ❤❤

[@kamelirzouni]

For the writing text, to avoid the extra comment from Claud 3.5 Sonnet, you need to specify it on artifacts mode

[@ebandaezembe7508]

Claude can do presentations in power point styles, just look at the demon presented by anthropic, and also I tested it and it works but for that you just have to ask to create a presentation not necessarily specifying powerpoint

[@amripfs92]

I am really frustrated with GPT-4o for coding, will try claude next!

[@ryan18462]

Sonnet 3.5 is such a beast

[@lhur23]

I believe those are not fake/invented source URLs. Many of the sites cited as source were valid at least up until the model was trained. For the complex reasoning test, most likely the models were trained on this specific riddles (or at least close variants) so may not be as valid. Perhaps one should come up with unique riddles for a good reasoning test.

[@shubharthaksangharsha6248]

You can actually can create PPTs in claude 3.5 sonnet by instructing it to create react based PPT (although it cant create tht .ppt ) but still very useful

[@newfrontiers5673]

Idk what you folks are having problems coding with 4o for. I litterally coded a webscraper in 4 python modules with a UI and only once did I have to paste in an error code.

[@andyt1313]

Thank you. ALL the other AI content channels are doing absolutely nothing but regurgitating the Claude demos. They add no value.

[@SkillLeapAI]

Thanks. I'm trying to just focus on value, examples and practical use cases as much as I can. too many channels have the AI news angle convered.

[@bigglyguy8429]

You can build custom GPTs with the free version I heard?

[@SkillLeapAI]

No you can’t build them. But you can use them

[@PseudoProphet]

Sam ClosedAI Man was trying to hide GPT-5 from us. Now he'll be forced to release it. 😂😂

[@thomassjovy4194]

its all good to use claude for smaller toy problems, but if you try to apply it professionally, you quickly realize it is just that - a toy for toy problems!

[@callibor3119]

Claude looks to be open source ready. Claude needs to be behind the back end to every other language model. Claude is the only successing language model out there and the most secured than all other language model. The sooner Claude becomes a back end to Apple, Microsoft, Google, Meta and Sony, the sooner we can actually have a safer digital space for WEB 3.0. This all needs to happen before November.

[@swoletech5958]

OpenAI will definitely drop something within a month to retake the lead.

[@phoebebrown2440]

Thanks for the breakdown. I agree, I use perplexity for research, which produces more realistic and real sources.

[@techikansh]

There should have been more coding comparison tasks in this video.

[@hitmusicworldwide]

I still had to get GPT 4o to rescue a coding project that Sonnet was starting to f up. So you can't go with just 1 model. Gemini 1.5 pro is a wreck

[@airevolution23]

You don’t have the ChatGPT Mac app?

[@SkillLeapAI]

Yea I have it. I mentioned it later in the video.

[@mcjorshs5]

Chatgpt 4o For programming, which one is better than this is good. I read them for programmers.

[@SkillLeapAI]

Claude is better than

[@mcjorshs5]

@@SkillLeapAI THANK YOU UNFORTUNATELY IT IS NOT AVAILABLE FOR MY COUNTRY USE VPN BUT IT WON'T LET YOU PAY

[@sdhpCH]

I second the conclusion.

[@issiewizzie]

This is more like it ...thanks

[@aiamfree]

gpt3.5 with code interpreter is GOAT.

[@wartem]

Claude seems superior in logical reasoning, while GPT-4o seems to give worse responses than its predecessors.

[@Dron008]

I didn't know ChatGPT is able to create presentations without gpts.

[@gorillatrail]

If you need better stories and coding go claude , if you want anything else its too controlled

[@DanandNato]

bekhoda to irani hasti. Hasti? Great channel btw :)

[@SkillLeapAI]

yes sir. thank you

[@hamedparsa8880]

الان یعنی ایرانیه؟ گفت یس سر! 😮😂

[@alonzoperez2470]

Chatgpt-4O wins because it sift through the web in real time

[@frustasistumbleguys4900]

Sonnet can create presentation slide using coding.

[@seventyfive7597]

Gpt 4o is clearly inferior to gpt 4 in math, rationality, deduction, code etc, it's clear for anyone who actually used it on hard problems. Oh and Sonnet 3.5 is clearly below both. Again, only if tested on uncontaminated questions. So no ridiculous snake programming etc

[@Hasmatix]

chatgpt is unlimited, int he meanwhile claude makes u water 5 hours after few messages

[@Dave-cg9li]

You should have used something like MS Word to count the words. The language model obviously cannot do that reliably 😅

[@Derick99]

Can you give me your opinion on the best local csv documents? Want to work with like 100k parts but want it private

[@SkillLeapAI]

Most of my experience with open source private LLM has been with Llama. I have this tutorial on it. Not sure about the context window through. kzbin.info/www/bejne/bnKycnSInctkmtU

[@Derick99]

@SkillLeapAI thank you I will try this all though last time I used docker it installed Linux on one of the things and I couldn't get rid of dual boot Linux on my system But I will try this

[@Mede_N]

Llama has a context window of 8k. Locally, you can try Phi 3 (small or medium), or one of the many mistral variants. Usually they have around 100k token window sizes (mistral usually around 32k). Make sure that you have enough RAM and time on your hands since 100k tokens take a lot of resources to be processed.

[@Mohsen-i6j]

ممنون ازت عالی بود

[@HenrikPflanz-o4d]

Maybe it is just me, but I feel that the response quality from ChatGPT-4, and 4o has degraded, and very good. I even prefer Llama 3 70B over ChatGPT at this point.

[@ryugurena3327]

No gpt is still the best one i've used

[@henrythegreatamerican8136]

Both of them hallucinate and both of them are overly censored. So I vote NONE OF THE ABOVE.

[@wheeltech242]

yeah brother great

[@JaiS0N]

Seems like GPT fails at promised functions where Claude didn’t really have that problem - Claude didn’t pretend to have a function that didn’t work like GPT. Also, GPT boldly provided false data in its responses. To me that is a huge fail. Web browsing: copy and paste web pages like you did. Neither are going to scrape pages well. And GPT failed to provide working links. I think the bottom line is that if you’re wanting to use either model with API, then Claude seems like the no-brainer. Claude seems more trustworthy.

[@saro.saribekyan]

Kind of a misleading title, you are more comparing the final applications built using the models, but not the actual models.

[@SkillLeapAI]

How else would you compare the two?

[@SkillLeapAI]

Claude and ChatGPT is how 99% of people interact with these models. I made the video for those people, not developers

[@saro.saribekyan]

@@SkillLeapAI thank you for the reply. I'm totally fine with that kind of a comparison covering the common use case. My concern was with the naming, as the video compared ChatGPT with Claude chat, but not the pure models.

[@Yipper64]

Seriously if Claude just stole a few OpenAI-isms with the GPTs and the memory, it would hands down beat out Chat GPT on every level. For my needs however, I think Claude is more useful to me currently despite all that. You said chat GPT and Claude are about even when it comes to visualizations, but I have to disagree. Claude is much better.

[@executiveproducer]

GTP 4o is bad compared to GPT 4. Why didn’t you use GPT 4 for the sake of comparison?

[@hodquan9430]

man i feel outdated just discovered sonnet not long ago its so better than gpt in coding

[@Prabawa-ig7dm]

You could try via poe it better.

[@itsjustme373]

Did you test claude’s voice feature? Oh, never mind.

[@froydinger]

This video is it for me. Bye GPT.

[@ChristiaanRoest79]

I have used chatgpt (3/4/5/4o) and i think its crap for analyzing legal text and writing text/letters. I have to adjust a lot before the grammar is correct. Claude Opus r0x. Opus is my favorite.

[@SkillLeapAI]

Is Sonnet beating Opus for you?

[@24-7gpts]

Bro you have ChatGPT 5?

[@HenrikPflanz-o4d]

@@24-7gpts probably a typo

[@ChristiaanRoest79]

@@SkillLeapAI At the moment i prefer Opus over Sonnet. I have noticed that the text generated by Opus for my legal projects is more accurate. But i have to experiment more to be absolute sure.

[@_.Madness._]

ChatGPT 5 seems to be the best for coding. The only problem is some days it refuses to debug unless you compliment its syntax first.

[@MrlegendOr]

This guy work for OpenAI or Microsoft ?🙃

[@SkillLeapAI]

lol I’m pretty sure I declared Claude the winner

[@ModernCentrist]

Claude is obviously better

[@picksilm]

powerpoint is microsoft, of course it works with openAI

[@morespinach9832]

Neither. I want locally hostable open source models that do small tasks well. And I’ve found that in mistral, lang chain and phi-3.

[@AtariWow]

Llama 3 is also good. You probably already know but the Ollama program makes running these models super easy. (Partially leaving this comment for others that may not know)

[@angloland4539]

❤

[@naturallydope247]

ChatGPT is horrible with links. It’s been that way since last year. Claude is much better for links.

[@angel_machariel]

Tongues claim 3.5 sonnet is far better with code. Should have demonstrated that as well.

[@MadaTangdilallo]

Not thing

[@aidajam3294]

It is garbage. I have really long experience using different GPTs (in programming). Claude (at least Sonnet) is making up stuff, therefore you should pay more attention how precise it is. In the end you will spend more time. On contrary ChatGPT (4o) is verbosy (even with your custrom instruction) but more accurate. IMHO, it is better to get annoying verbosy results instead of keeping constantly in eye on result's precison.

[@PerfectArmonic]

Both of them are disappointing and doesn’t deserve to pay for them. They should be able to ask questions and to learn from them. As long as they answer with points and repeating half of my prompt they are not intelligent at all. And they are limited in a lot of domains.

[@bag1667]

Your mom doesn't deserve a tip pay