Finally a Competitor! Claude 3.5 Sonnet vs Deepseek v3 Who will Win?

Поділитися
Вставка
  • Опубліковано 23 січ 2025

КОМЕНТАРІ • 54

  • @TheDiamondHawkOfficial
    @TheDiamondHawkOfficial 26 днів тому +13

    I was surprised when I say you have less than 500 subs. Your tests are so good i have notification on so I can watch them. Super underrated!!!!!

    • @YJxAI
      @YJxAI  26 днів тому +1

      🥹 thank you very much

    • @michaelspoden1694
      @michaelspoden1694 26 днів тому +3

      Remember me when you get famous😂

    • @michaelspoden1694
      @michaelspoden1694 26 днів тому +2

      I was talking to the 3 Model early this morning and 37 billion per expert that's a lot of experts for a 671 billion parameter model. Obviously it will use multiple experts.
      You should compare the Claude Sonnet 3.5 versus the open router Claude sonnet 3.5. At least with other models I am definitely seen the app versus Google AI Studio or other API or UI of the same model be completely noticeably different. Like the Google is a studio does not have the roadblocks does the app does not even close and that's without tweaking the temperature, tools user instructions anything. I just think it would be cool to see if you notice significantly different capabilities.

    • @YJxAI
      @YJxAI  26 днів тому +1

      @@michaelspoden1694

    • @YJxAI
      @YJxAI  26 днів тому +2

      @@michaelspoden1694 i have noticed difference in my case it was gpt4o in the api or gpt4o i the chatgpt. But now they have chatgpt latest in the api .
      But that is indeed a good idea of comaparrison.

  • @wwkk4964
    @wwkk4964 26 днів тому +1

    Great review!

    • @YJxAI
      @YJxAI  25 днів тому

      thanks

  • @rcingfever3882
    @rcingfever3882 26 днів тому +5

    Finally some one beats it.

  • @SpragginsDesigns
    @SpragginsDesigns 24 дні тому +2

    I subscribed. Other videos seem to be catching onto DeepSeek, but their videos just dont really cover it well for coding especially. Or the DeepThink reasoning. I would say most developer's flows now are in Cursor. So, utilizing it in that or Cline or Windsurf etc would be a great test of coding abilities.

    • @YJxAI
      @YJxAI  24 дні тому +1

      good idea

    • @SpragginsDesigns
      @SpragginsDesigns 24 дні тому

      @YJxAI thanks! I definitely am using DeepSeek instead of ChatGPT now. It's free and faster. So the $20/month plan just seems like pointless for me. And I use Sonnet for coding tasks. And it has reasoning, so it seems to sort of negate the need for chatgpt plus.

  • @aniketsahoo5061
    @aniketsahoo5061 26 днів тому +2

    Good job broo

  • @npc4416
    @npc4416 26 днів тому +3

    Very good channel

    • @YJxAI
      @YJxAI  26 днів тому +1

      🥹

  • @Samt2b
    @Samt2b 25 днів тому +1

    Whats the link that you test the codes

    • @YJxAI
      @YJxAI  25 днів тому

      openrouter.ai .
      If you are talking about the leaderboard and all that was my local server .

  • @adarshraj5167
    @adarshraj5167 26 днів тому +1

    can you compare deepseek and claude with gpt o1?

    • @YJxAI
      @YJxAI  25 днів тому

      will do

  • @RickySupriyadi
    @RickySupriyadi 26 днів тому +2

    how good it will be when running in agentic mode

    • @YJxAI
      @YJxAI  26 днів тому +1

      yeah that is also good to see how good it is wen it comes to instrction follwing and givng out proper outputs but that kinda will be to narrow domain.

    • @gijrijasdas5055
      @gijrijasdas5055 26 днів тому +1

      i made one with custom RAG , very incridible result (comparing API SUPER DUPER EXTRA HOLLYY CHEAP)

    • @RickySupriyadi
      @RickySupriyadi 26 днів тому

      @@gijrijasdas5055 cool

    • @YJxAI
      @YJxAI  26 днів тому +1

      @@gijrijasdas5055 yeah i am planning to try it out i wish i could use it in windsurf.

  • @とふこ
    @とふこ 26 днів тому +2

    Deepseek are smarter especially if i turn on the thinking, but Claude have better writing style (in Hungarian language) and better mobile app.

    • @YJxAI
      @YJxAI  26 днів тому

      ohh i see

  • @Valk-Kilmer
    @Valk-Kilmer 26 днів тому +1

    How is it a fail because you don't have the proper Python module installed? That's on you, not Claude.

    • @YJxAI
      @YJxAI  25 днів тому

      Yeah I understand brother what you are saying. I also thought that was the case but it's actually an import error cause there is no module named pyqtSignal. So even when I tried to correct the import error even after that it didn't work.
      IT is genuine doubt that could arise from anyone and it's okay.

    • @Valk-Kilmer
      @Valk-Kilmer 25 днів тому +1

      @@YJxAI Signal is part of QT which is a UI framework for Python. So you need QT Framework installed to use it. So I wouldn't say Claude failed just that it requires QT to be used properly.

  • @Vinayone8
    @Vinayone8 26 днів тому +1

    Deepseek is gpt wrapper

    • @anandkanade9500
      @anandkanade9500 26 днів тому +6

      but open source , so ....

    • @hareram4233
      @hareram4233 26 днів тому

      How do u know it's gpt wrapper

    • @MikeAlan-d4t
      @MikeAlan-d4t 26 днів тому

      Oh, just secretly rejoice then! Offering services at one-tenth of the price. What a "bargain"!🤭

    • @huk2617
      @huk2617 25 днів тому

      it's not

    • @gene61ify
      @gene61ify 25 днів тому +1

      but it is cheaper???????????????

  • @HimaLoubi
    @HimaLoubi 25 днів тому +1

    Deepseek is sick 😅 not even close to Claude haiku

  • @nawaz_haider
    @nawaz_haider 26 днів тому +1

    Actually

  • @UsmanAli-ve6tq
    @UsmanAli-ve6tq 25 днів тому +1

    According to your results gpt 4o mini is better than gpt 4o. This makes no sense.

    • @FalconStudioWin
      @FalconStudioWin 25 днів тому +1

      It's actually possible true in some cases from my own personal testing I am not sure why that's the case but overall o4 is better but fails in easy questions as mini passes

    • @YJxAI
      @YJxAI  25 днів тому +2

      out of the small models my faviourite is gpt4o mini . It is intelligence in pennies. as it was shown in the video I used 800k tokens and was prices somewhere around 0.02 dollars or something I mean WTF! 😂

  • @remsee1608
    @remsee1608 26 днів тому +1

    The model rules. Most importantly I don’t get woke lectures from it 😂

  • @AnonymousObject
    @AnonymousObject 25 днів тому +1

    is this English

    • @huk2617
      @huk2617 25 днів тому +1

      yes, yours must be quite bad if you can't understand it just because he has an accent

    • @AnonymousObject
      @AnonymousObject 25 днів тому +1

      @@huk2617 I can't understand any word he said.

    • @YJxAI
      @YJxAI  25 днів тому

      Thanks for understanding

    • @YJxAI
      @YJxAI  25 днів тому +1

      😅 yeah i know i am working on it