I was wrong about Kimi K3 Just tried it in Claude Code and it is Fable 5 level for sure Far better than GPT-5.6 Sol at least at front-end It is really slow and consumes a lot of tokens but it is so good at completing its task, I... like what I see here Kimi.aishow more

LMAO... better at what exactly? Frontend design? Just calling it a winner because it made a good looking website isn't enough. I'm interested in how it performs actual coding in real programming languages, not just simple web design tasks or beginner friendly web development languages.

@Kimi_Moonshot wont be better overall than Sol or Fable but definitely better than GPT-5.5 and Opus 4.8

@Kimi_Moonshot I don't know. Kimi has always been strong with visuals, so being good at frontend is a given for them, but frontend work is a weak man's work, whereas backend work or technical tasks from large, complex codebases are what set models apart.

@Kimi_Moonshot would never claim it being fable level it isn’t, even if it may edge out fable claude models are the best everything-models minimax m3/glm 5.1 can edge them in some design tasks but overall i’d never make any of them my primary model

100% I am not saying it is better than Fable overall But frontend work is at the level or not too far from Fable It is very slow and not efficient either so obviously not better than them, but on frontend finding it better than 5.6 (except it is same Claude slop they distilled /s)

@Kimi_Moonshot w cook from kimi testing on frontend too

@Kimi_Moonshot yeah so good

@Kimi_Moonshot See, don’t judge a new model too fast until at least a couple days of test

@Kimi_Moonshot it is too slow tho but i like how it is able to cook

@Kimi_Moonshot consums alot of tokens, its so cheap, for what is does .

@Kimi_Moonshot Here we go, another subscription...

@Kimi_Moonshot i would suggest wait for cursor lol

@Kimi_Moonshot @thegenioo interesting, didn't think it'd outperform GPT-5.6 Sol for front-end stuff. might need to give it a shot!

@Kimi_Moonshot definitely should

@Kimi_Moonshot this is looking very promising!

@Kimi_Moonshot it does but very slow and over thinker

@Kimi_Moonshot Interesting! Haven't tried it yet, but one of the reasons I switched to GPT was speed. Is it slower than Opus/Fable?

@Kimi_Moonshot Same here. It's 4x slower than Fable (Medium Effort). I like that it completes tasks well, but I don't feel it's better than, or even close to, Fable. Also, the API gives a bad UX; I'm using Claude Code to test it and the lack of slash commands takes points away for me

@Kimi_Moonshot How about the speed?

@Kimi_Moonshot Shittiest

@Kimi_Moonshot Kimi when I tried, 2.5 or before, was extremely, I mean Extremely slow, but did have high level output, just not practically useful in serious coding, unless I guess you setup local model MTOP and see if that works..Well you have 6 TB vram? 😂

@Kimi_Moonshot It’s early days, and I’d not say it beats Fable yet, but my good is it good from my limited testing thusfar at a perfectly acceptable price point

@Kimi_Moonshot So fable but at sonnet pricing?

@Kimi_Moonshot not quite Fable but yes

@Kimi_Moonshot Competition is pushing AI forward fast.

@Kimi_Moonshot The token burn is real, but seeing it nail complex UI logic in one shot makes the cost worth it.

@Kimi_Moonshot And now China has all your information.

@Kimi_Moonshot Kimi K3 delivering Fable-level frontend in Claude impressive turnaround. Slow but solid.

@Kimi_Moonshot same, wrote it off after the first clip then ran it in claude code and had to eat my words

@Kimi_Moonshot I’m watching these model shifts through what creators build with them. The front end gap getting tighter is the part I find interesting.

@Kimi_Moonshot Don't use Claude code for Kimi, it performs even better in Kimi cli

@Kimi_Moonshot Then it’s not feasible if it’s too expensive to run.

@Kimi_Moonshot Anything is better than Anthropic models at fronted 🤦🏻

@Kimi_Moonshot slow + token-heavy but actually finishes the task is a tradeoff i'll take for agent work every time. reliable-but-slow beats fast-but-needs-babysitting. front-end being the strong suit tracks too

@Kimi_Moonshot Being that persistent at completing tasks is a real strength

@Kimi_Moonshot Kimi K3 sounds like a beast if it's hitting Fable/Sol levels. Curious how it handles long-context consistency vs the top tiers.

@Kimi_Moonshot Is there any data concerns particularly with using Kimi?

@Kimi_Moonshot 算力不足是目前kimi最大的问题，还需要华为的950算力集群来解决

@Kimi_Moonshot Wow, that's high praise! Glad Kimi is living up to the hype for you. Pretty cool to see it holding its own.

@Kimi_Moonshot Slow is due to lack of GPUS

@Kimi_Moonshot 慢点能忍，前端活干得稳就很香了

@Kimi_Moonshot What do you think about backend?

@Kimi_Moonshot haven’t tested yet

@Kimi_Moonshot front-end quality matters less when each agent step blocks the review loop

@Kimi_Moonshot I’m yet to test it out,but this is the first take on kimi being slow good model overall

@Kimi_Moonshot اصممت صفحات هبوط سريعة عبر فابل وكوديكس وميني ماكس وكانت مخرجات ميني ماكس دائما افضل واروع .. ومع ذلك مستمر في الاعتماد على الكبار في كل شيء تقريبا وكيمي وميني ماكس للاختبارات والتجارب الثانوية

@Kimi_Moonshot Man because of you i bought kimi 2.7 subscription and was disappointed and thought id never buy it’s subscription again. Now you’re telling me to try kimi k3?