Kimi K3 still trails Anthropic’s Claude Fable 5 and OpenAI’s GPT 5.6 Sol on overall performance, the company said on Friday, but consistently outperformed other tested models.

The model beat Claude Opus 4.8 and GPT 5.5 — models that sit just behind Anthropic and OpenAI’s leading-edge systems — on benchmarks including coding and general agents, according to Moonshot.

It’s China’s largest AI model so far, with 2.8 trillion parameters, referring to the size of its neural network.

  • AceFuzzLord@lemmy.zip
    link
    fedilink
    English
    arrow-up
    4
    ·
    24 hours ago

    Who xares whether it’s better or not. A dumpster fire is still a dumlster fire no matter who made it or how big it is.

    • chocrates@piefed.world
      link
      fedilink
      English
      arrow-up
      9
      ·
      2 days ago

      How big is un-quantized 2.8t param model?

      I just got some hardware but I still don’t have a ton of vram :(

      • SuspiciousCarrot78@aussie.zone
        link
        fedilink
        English
        arrow-up
        5
        ·
        20 hours ago

        It’s one of those “if you have to ask, you can’t afford it” scenarios.

        I’ve heard FP16 is likely to be 6TB of VRAM, when Kvcache is factored in.

        Even a more sensible quant will likely be 1TB.

        … most of us are not running this at home.

          • SuspiciousCarrot78@aussie.zone
            link
            fedilink
            English
            arrow-up
            2
            ·
            4 hours ago

            I was looking thru an old box of bits yesterday for a project - I needed 2x8GB sticks of ddr3, bought circa 2023. Lo and behold, they had sent me 2x16 DDR4. Wrong size, wrong type, wrong timing.

            I’d be mad but apparantly I’m now the king of Londinium and can afford a shiny hat.

      • L_Acacia@lemmy.ml
        link
        fedilink
        English
        arrow-up
        19
        ·
        2 days ago

        Its trained at native MXFP4 with MXFP8 activations layer, so you need around 1.5TB of VRAM to fully offload it without taking into account the context cache. It might be doable to do some smart expert offloading and swapping, but expect minimum 500GB of VRAM and 1TB of system RAM minimum and the t/s would be reduced.

        Its only realistically runnable on datacenter grade gpu at decent speed for now (and judging from the price of ram for the next few years)

      • cecilkorik@lemmy.ca
        link
        fedilink
        English
        arrow-up
        10
        ·
        2 days ago

        Hobbyists are working in the billions or tens of billions. Trillions is insane. I’m sure some people can run it, but not many.

        • partofthevoice@lemmy.zip
          link
          fedilink
          English
          arrow-up
          4
          arrow-down
          1
          ·
          1 day ago

          I know a guy who found a way to monetize his GPU, just really slowly. He did it until he was able to buy 2 more, then 4 more, and so on. He has like 90 now. I’m sure he could if he wanted to.

  • A_A@lemmy.world
    link
    fedilink
    English
    arrow-up
    6
    ·
    2 days ago

    if it wasn’t for the humongous money bubble, the dreadful Terminator Horizon, the catastrophic climate effects and so on, one could say these tools are “valuable”.

    • HieroProtagonist@lemmy.ml
      link
      fedilink
      English
      arrow-up
      4
      arrow-down
      4
      ·
      2 days ago

      The money bubble? Will correct itself soon

      The Terminator thing? Is bullshit - humans are already very good at terminating each-other

      The catastrophic climate effects? - In relation to other factors like meat consumption auiet neglitabile

      • A_A@lemmy.world
        link
        fedilink
        English
        arrow-up
        3
        ·
        1 day ago

        Meat : you are right,
        Terminator : not the humanoid form and no time travel … but “a.i.” is already in drones,
        Bubble : will burst and many ill advised folks will lose.