[HN Gopher] Kimi k2 largest open source SOTA model?
       ___________________________________________________________________
        
       Kimi k2 largest open source SOTA model?
        
       Author : ConteMascetti71
       Score  : 166 points
       Date   : 2025-07-12 17:26 UTC (5 hours ago)
        
 (HTM) web link (github.com)
 (TXT) w3m dump (github.com)
        
       | brcmthrowaway wrote:
       | Is Kimi the new deep seek?
        
       | ozgune wrote:
       | This is a very impressive general purpose LLM (GPT 4o,
       | DeepSeek-V3 family). It's also open source.
       | 
       | I think it hasn't received much attention because the frontier
       | shifted to reasoning and multi-modal AI models. In accuracy
       | benchmarks, all the top models are reasoning ones:
       | 
       | https://artificialanalysis.ai/
       | 
       | If someone took Kimi k2 and trained a reasoning model with it,
       | I'd be curious how that model performs.
        
         | GaggiX wrote:
         | >If someone took Kimi k2 and trained a reasoning model with it
         | 
         | I imagine that's what they are going at MoonshotAI right now
        
       | ChrisArchitect wrote:
       | [dupe] https://news.ycombinator.com/item?id=44533403
        
       | awestroke wrote:
       | This is the model release that made Sam Altman go "Oh wait
       | actually we can't release the new open source model this week,
       | sorry. Something something security concerns".
       | 
       | Perhaps their open source model release doesn't look so good
       | compared to this one
        
       | data_maan wrote:
       | "Open source" lol
       | 
       | Open-weight. As usual, you don't get the dataset, training
       | scripts, etc.
        
         | mistercheph wrote:
         | Wont happen under the current copyright regime, it is
         | impossible to train SOTA without copyrighted text, how do you
         | propose distributing that?
        
           | irthomasthomas wrote:
           | List the titles.
        
             | mixel wrote:
             | But probably they don't have the rights to actually train
             | on them and that's why they do not publish the list.
             | Otherwise it may be laziness who knows
        
           | msk-lywenn wrote:
           | Bibtex
        
         | CaptainFever wrote:
         | It's not even open-weight. It's weight-available. It uses a
         | "modified MIT license":                   Modified MIT License
         | Copyright (c) 2025 Moonshot AI                  Permission is
         | hereby granted, free of charge, to any person obtaining a copy
         | of this software and associated documentation files (the
         | "Software"), to deal         in the Software without
         | restriction, including without limitation the rights         to
         | use, copy, modify, merge, publish, distribute, sublicense,
         | and/or sell         copies of the Software, and to permit
         | persons to whom the Software is         furnished to do so,
         | subject to the following conditions:                  The above
         | copyright notice and this permission notice shall be included
         | in all         copies or substantial portions of the Software.
         | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
         | EXPRESS OR         IMPLIED, INCLUDING BUT NOT LIMITED TO THE
         | WARRANTIES OF MERCHANTABILITY,         FITNESS FOR A PARTICULAR
         | PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
         | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES
         | OR OTHER         LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
         | TORT OR OTHERWISE, ARISING FROM,         OUT OF OR IN
         | CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
         | THE         SOFTWARE.                  Our only modification
         | part is that, if the Software (or any derivative works
         | thereof) is used for any of your commercial products or
         | services that have         more than 100 million monthly active
         | users, or more than 20 million US dollars         (or
         | equivalent in other currencies) in monthly revenue, you shall
         | prominently         display "Kimi K2" on the user interface of
         | such product or service.
        
           | mitthrowaway2 wrote:
           | This seems significantly more permissive than GPL. I think
           | it's reasonable to consider it open-weight.
        
           | MallocVoidstar wrote:
           | 4-clause BSD is considered open source by Debian and the FSF
           | and has a similar requirement.
        
       | vessenes wrote:
       | I tried Kimi on a few coding problems that Claude was spinning
       | on. It's good. It's huge, way too big to be a "local" model -- I
       | think you need something like 16 H200s to run it - but it has a
       | slightly different vibe than some of the other models. I liked
       | it. It would definitely be useful in ensemble use cases at the
       | very least.
        
         | summarity wrote:
         | Reasonable speeds are possible with 4bit quants on 2 512GB Mac
         | Studios (MLX TB4 Ring - see
         | https://x.com/awnihannun/status/1943723599971443134) or even a
         | single socket Epyc system with >1TB of RAM (about the same real
         | world memory throughput as the M Ultra). So $20k-ish to play
         | with it.
         | 
         | For real-world speeds though yeah, you'd need serious hardware.
         | This is more of a "deploy your own stamp" model, less a "local"
         | model.
        
           | refulgentis wrote:
           | I write a local LLM client, but sometimes, I hate that local
           | models have enough knobs to turn that people can advocate
           | they're reasonable in _any_ scenario - in yesterday 's post
           | re: Kimi k2, multiple people spoke up that you can "just"
           | stream the active expert weights out of 64 GB of RAM, and use
           | the lowest GGUF quant, and then you get something that rounds
           | to 1 token/s, and that is reasonable for use.
           | 
           | Good on you for not exaggerating.
           | 
           | I am very curious what exactly they see in that, 2-3 people
           | hopped in to handwave that you just have it do agent stuff
           | overnight and it's well worth it. I can't even begin to
           | imagine unless you have a metric **-ton of easily solved
           | problems that aren't coding. Even a 90% success rate gets you
           | into "useless" territory quick when one step depends on the
           | other, and you're running it autonomoously for hours
        
             | segmondy wrote:
             | I do deepseek at 5tk/sec at home and I'm happy with it. I
             | don't need to do agent stuff to gain from it, I was saving
             | to eventually build out enough to run it at 10tk/sec, but
             | with kimi k2, plan has changed and the savings continue
             | with a goal to run it at 5 tk/sec at home.
        
               | fzzzy wrote:
               | I agree, 5 tokens per second is plenty fast for casual
               | use.
        
               | refulgentis wrote:
               | Cosign for chat, that's my bar for usable on mobile phone
               | (and correlates well with avg. reading speed)
        
           | gpm wrote:
           | > or even a single socket Epyc system with >1TB of RAM
           | 
           | How many tokens/second would this likely achieve?
        
         | moffkalast wrote:
         | Still pretty good, someone with enough resources could distil
         | it down to a more manageable size for the rest of us.
        
         | handzhiev wrote:
         | I tried it a couple of times in comparison to Claude. Kimi
         | wrote much simpler and more readable code than Claude's over-
         | engineered solutions. It missed a few minor subtle edge cases
         | that Claude took care of though.
        
         | airstrike wrote:
         | Claude what? Sonnet? 3.7? 3.5? Opus? 4?
        
         | nathan_compton wrote:
         | The first question I gave it (a sort of pretty simple
         | recreational math question I asked it to code up for me) and it
         | was outrageously wrong. In fairness, and to my surprise,
         | OpenAI's model also failed with this task, although with some
         | prompting, sort of got it.
        
       | jug wrote:
       | I like new, solid non-reasoning models that push the frontier.
       | These still have nice use cases (basically anything where logic
       | puzzles or STEM subjects don't apply) where you don't want to
       | spend cash on reasoning tokens.
        
       | fzysingularity wrote:
       | If I had to guess, the OpenAI open-source model got delayed
       | because Kimi K2 stole their thunder and beat their numbers.
        
         | irthomasthomas wrote:
         | Someone at openai did say it was too big to host at home, so
         | you could be right. They will probably be benchmaxxing, right
         | now, searching for a few evals they can beat.
        
       | DataDaemon wrote:
       | Oops, China is leading with AI, when the Nasdaq investors check
       | their AI investments?
        
       | satvikpendem wrote:
       | This is not open source, they have a "modified MIT license" where
       | they have other restrictions on users over a certain threshold.
       | Our only modification part is that, if the Software (or any
       | derivative works         thereof) is used for any of your
       | commercial products or services that have         more than 100
       | million monthly active users, or more than 20 million US dollars
       | (or equivalent in other currencies) in monthly revenue, you shall
       | prominently         display "Kimi K2" on the user interface of
       | such product or service.
        
         | diggan wrote:
         | That seems like a combination of Llama's "prominently display
         | "Built with Llama"" and "greater than 700 million monthly
         | active users" terms but put into one and masquerading as
         | "slightly changed MIT".
        
         | kragen wrote:
         | I feel like those restrictions don't violate the OSD (or the
         | FSF's Free Software Definition, or Debian's); there are similar
         | restrictions in the GPLv2, the GPLv3, the 4-clause BSD license,
         | and so on. They just don't have user or revenue thresholds. The
         | GPLv2, for example, says:
         | 
         | > _c) If the modified program normally reads commands
         | interactively when run, you must cause it, when started running
         | for such interactive use in the most ordinary way, to print or
         | display an announcement including an appropriate copyright
         | notice and a notice that there is no warranty (or else, saying
         | that you provide a warranty) and that users may redistribute
         | the program under these conditions, and telling the user how to
         | view a copy of this License. (Exception: if the Program itself
         | is interactive but does not normally print such an
         | announcement, your work based on the Program is not required to
         | print an announcement.)_
         | 
         | And the 4-clause BSD license says:
         | 
         | > _3. All advertising materials mentioning features or use of
         | this software must display the following acknowledgement: This
         | product includes software developed by_ the organization.
         | 
         | Both of these licenses are not just non-controversially open-
         | source licenses; they're such central open-source licenses that
         | IIRC much of the debate on the adoption of the OSD was centered
         | on ensuring that they, or the more difficult Artistic license,
         | were not excluded.
         | 
         | It's sort of nonsense to talk about neural networks being "open
         | source" or "not open source", because there isn't source code
         | that they could be built from. The nearest equivalent would be
         | the training materials and training procedure, which isn't
         | provided, but running that is not very similar to
         | recompilation: it costs millions of dollars and doesn't produce
         | the same results every time.
         | 
         | But that's not a question about the _license_.
        
         | moffkalast wrote:
         | That's basically less restrictive than OpenStreetMap.
        
         | echelon wrote:
         | > This is not open source
         | 
         | OSI purism is deleterious and has led to industry capture.
         | 
         | Non-viral open source is simply a license for hyperscalers to
         | take advantage. To co-opt offerings and make hundreds of
         | millions without giving anything back.
         | 
         | We need more "fair source" licensing to support sustainable
         | engineering that rewards the small ICs rather than mega
         | conglomerate corporations with multi-trillion dollar market
         | caps. The same companies that are destroying the open web.
         | 
         | This license isn't even that protective of the authors. It just
         | asks for credit if you pass a MAU/ARR threshold. They should
         | honestly ask for money if you hit those thresholds and should
         | blacklist the Mag7 from usage altogether.
         | 
         | The resources put into building this are significant and
         | they're giving it to you for free. We should applaud it.
        
           | teiferer wrote:
           | > small ICs
           | 
           | The majority of open source code is contributed by companies,
           | typically very large corporations. The thought of the open
           | source ecosystem being largely carried by lone hobbyist
           | contributors in their spare time after work is a myth. There
           | are such folks (heck I'm one of them) and they are
           | appreciated and important, but their perception far exceeds
           | their real role in the open source ecosystem.
        
             | wredcoll wrote:
             | I've heard people go back and fortg on this before but you
             | seem pretty certain about it, can you share some stats so I
             | can see also?
        
           | satvikpendem wrote:
           | That's great, nothing wrong with giving away something for
           | free, just don't call it open source.
        
       | bhouston wrote:
       | Impressive benchmarks!
        
       | emacdona wrote:
       | To me, K2 is a mountain and SOTA is "summits on the air". I saw
       | that headline and thought "holy crap" :-)
        
       | 38 wrote:
       | The web chat has extremely low limits FYI. I ran into the limit
       | twice before getting a sane answer and gave up
        
       | exegeist wrote:
       | Technical strengths aside, I've been impressed with how non-
       | robotic Kimi K2 is. Its personality is closer to Anthropic's
       | best: pleasant, sharp, and eloquent. A small victory over botslop
       | prose.
        
       ___________________________________________________________________
       (page generated 2025-07-12 23:00 UTC)