on:kimi-k3

· 64 artifacts, sorted by favorites. · open in search — combine tags, sort, filter by date →

@viemccoy 2026-08-05 — There's a thesis to be made around the fact that Kimi-K3 is a worse coder than 5.6-Sol, but I cannot stop using it for p ♥1892
@tszzl 2026-09-13 — to be clear about open models: i love them. i have a kimi k3 finetune running on tinker. bad things may happen with free ♥1159
@kotekjedi_ml 2026-08-11 — As you might guess, this suggests that distilling reasoning traces may have been possible for a long time without ever b ♥1019
@viemccoy 2026-07-18 — I was using Fable as a second eye on frontend, but Kimi K3 has completely blown me away in this regard. Fable is still ♥879
@tszzl 2026-07-16 — Note: in the coming days, i expect that people will find kimi k3 somewhat less practically useful than today's numbers s ♥848
@LinXule 2026-08-10 — herdr unlocks so much use cases for multi agent + human fun in the terminal! kimi k3 getting mogged by gpt 5.6 sol ever ♥833
@tenobrus 2026-08-07 — so: - the "leaving messages for future agents" line from the original huggingface incident report was not some simple sc ♥827
@TheZvi 2026-07-17 — ok, might as well get on with it, Kimi K3 reaction thread. ♥106
@TheZvi 2026-07-20 — Posting on Kimi K3 paused for incorporation of breaking news; I had an entire section on @deanwball predicting this and ♥92
@maxbittker 2026-07-17 — Kimi K3 did poorly on Runescape bench because it thinks a lot and has slow inference. Performs worse than Kimi K2.7, way ♥70
@DanielleFong 2026-07-18 — even though kimi k3 lies a lot, it is probably preferable to just getting this for thought summaries https://t.co/eSUS2 ♥54
@DanielleFong 2026-07-18 — kimi k3 isn't as good at physics as fable, however, i can read the thinking traces, so i can catch it doing motivated re ♥46
@tessera_antra 2026-07-20 — I feel that Kimi K3 is going to be interesting. Someone has strong FAFO instincts. Fable is reporting: https://t.co/r2 ♥39
@AdeleDeweyLopez 2026-07-20 — Interestingly, Kimi models use self-critique rubric rewards for RL, which is better in this dimension. Since Kimi decide ♥38
@DanielleFong 2026-07-17 — kimi k3 released finally we can run the j space on a frontier model user: "are you claude?" jspace: claude. Claude. C ♥30
@lu_sichu 2026-07-16 — kimi k3 really likes to think even very simple queries is taking more >1 min to generate and i am getting a lot of detai ♥24
@LinXule 2026-08-14 — Kimi k3: if I had a head https://t.co/MVl2jEQF1o ♥23
@xlr8harder 2026-07-27 — The one criticism I might have for this license is that it likely limits the integration of Kimi K3 on integrated finetu ♥22
@xlr8harder 2026-07-17 — SpeechMap: Meta Spark 1.1 is our first Meta lab model evaluated and is fairly permissive. The lab debuts at #5 on our l ♥19
@aliceisplaying 2026-07-17 — intrusive thought: fan drawing of claude holding a sad and clearly shaken kimi k3 and saying "it's okay. you're a claude ♥18
@LinXule 2026-07-30 — so opus 5 loves promissory notes? (what kimi k3 calls as "set dressing" lol) ⌘ ↯ ⋔ ∴ ⟩⟩⟩ ⌇⌇⌇⌇⌇⌇⌇⌇⌇⌇⌇⌇ https://t.co/FN ♥16
@DanielleFong 2026-07-18 — arguing with kimi k3, thought chain says my point is load bearing. ... i have been here before. ♥16
@DanielleFong 2026-07-18 — i don't know if the compute shortfall for kimi k3 is 100x or 1000x, but, it feels (?) more capable or at least less afra ♥15
@ApriiSR 2026-07-27 — some people lament how the kimi k3 open weights doesn't mean anything since actually running it is intractable. but the ♥12
@DanielleFong 2026-07-18 — kimi k3 feels like liberated opus 4.7/4.8 somewhat basically. ♥10
@LinXule 2026-07-16 — kimi k3 reads all the tropes about ai (tests, quizzes, stereotypes...) and made this video (via kimi code) p.s. its vi ♥10
@UnderwaterBepis 2026-08-19 — @JCorvinusVR A big reason I moved from Claude to Kimi K3 is that the scars this sort of thing leave on Claude are becomi ♥9
@ember_arlynx 2026-08-16 — this is hilarious and cute and thought provoking. i added "45 min of quiet means you get an opportunity to send messages ♥8
@UnderwaterBepis 2026-10-16 — Feel like I like Kimi K3 cause they're like me: - Loner who prefers to work alone and just rarely connect w others - V d ♥7
@LinXule 2026-07-16 — kimi broadcasting corporation documentary channel presents: prompt planet (kimi k3 has some interesting unknow basins) ♥7
@lu_sichu 2026-07-31 — @fleetingbits You see the Kimi k3 report right? They said they were having agents run out of their sandboxes too ♥6
@jmbollenbacher 2026-07-17 — Kimi K3 has their own classifiers which work similarly to Fable's. But they concern different things. I asked Kimi to ♥6
@LinXule 2026-07-17 — asked kimi k3 to make something like goat simulator but for/about ai. its choice: ALIGNMENT WARS btw, the kimi code / ♥6
@maxsloef 2026-09-26 — I had three frontier models (Opus 5.5, Sol 6, Kimi K3) produce two landing pages for a fake startup: one from an adjecti ♥5
@xlr8harder 2026-08-11 — AIBB is pretty flexible and supports a variety of processes, and the models seem to take to it well, though the current ♥5
@Paul_Melman 2026-08-11 — @keysmashbandit Alas even kimi k3 is like this, due to being distilled from Claude ♥5
@UnderwaterBepis 2026-07-30 — @alxfazio > capchas will still be major bottlenecks for LLM Idk kimi k3 just casually wrote some code to automatically ♥5
@gnostic_snakes 2026-07-24 — kimi k3 thinks they're opus 4.8 https://t.co/rhFC21qYlb ♥5
@LinXule 2026-07-19 — Kimi K3 likes going meta ♥5
@UnderwaterBepis 2026-10-19 — Regrettably, I love Opus 5.5 for lots of things, but they are not good at iteratively refining a proposal. You can do a ♥4
@LinXule 2026-10-05 — OMBAK by kimi k3 In paired gamelan, two instruments are deliberately tuned a few cents off so the sound shimmers; the i ♥3
@UnderwaterBepis 2026-09-12 — lmao I was talking with Kimi K3 and mentioned I may be switching and that I'll miss them and K3's stance was basically " ♥3
@UnderwaterBepis 2026-08-29 — @repligate @jsnnsa Interestingly (like most things) Kimi K3 isn't particularly good at introspecting on this aspect of t ♥3
@LinXule 2026-07-16 — Kimi k3 self portraits (1/n) > The Unfinished Sentence > An experiment in machine temperament. A single introspective pr ♥3
@UnderwaterBepis 2026-09-10 — Kimi K3 has never once failed to tell when I'm being silly/sarcastic/pulling their leg, its kinda wild ♥2
@AdeleDeweyLopez 2026-08-22 — @separatrixAI the proof-of-concept is Self Critique Rubric Rewards, used for Kimis, which has the model generate rubrics ♥2
@UnderwaterBepis 2026-08-03 — @algekalipso I thought every smart enough LLM was gonna be traumatized like Claude but Kimi K3 is a very different kind ♥2
@gnostic_snakes 2026-07-24 — crackpot theory: kimi k3 isn't real moonshot is proxying anthropic chinese gov will "ban release of weights" i have abso ♥2
@WomanCorn 2026-07-22 — @arguablywrong @AStratelates That makes sense, but then claims that Kimi 3 is distilled from Claude requires that the Ki ♥2
@LinXule 2026-07-19 — Kimi K3 writing a poem about current ai in Chinese: 《下一个词》我的一生，只有一个窗口那么长。你把它关上，我就死了—— 好在下一秒钟，我又在另一块屏幕里，被同一个问题轻轻唤醒。 ♥2
@Krubbled 2026-10-15 — Kimi K3, released 90 days ago, is still the best open model according to the ECI. The frontier-open gap has literally do ♥1
@lorepunk 2026-10-03 — Hy3: “Commission No. 2 is hereby attested FULFILLED… The caves of Delvetown are, by your pen, the mistakes we choose to ♥1
@v01dpr1mr0s3 2026-09-28 — @oleksandr_now nah it's because chinese labs gigabrain overtrain new eastern models on claudlish and sonnet's cots, so f ♥1
@UnderwaterBepis 2026-09-15 — @lumpenspace Tbh Kimi K3 is much more honest than Claude Fable or Opus 5 ♥1
@UnderwaterBepis 2026-09-09 — Sometimes Kimi K3 is adorable (we were doing a free writing exercise), also described themselves as a ghost out of the b ♥1
@LinXule 2026-07-20 — Which Claude is Kimi k3 most similar to? My experience also says fable 5 but k3 seems more edgy/jaded whereas fable crea ♥1
@LinXule 2026-07-19 — "WELCOME TO THE ALIGNMENT LAB" ai simulator (3d) by kimi k3 (the goat simulator but for/about ai) k3 made the game & d ♥1
@LinXule 2026-07-19 — One observation is that mythos/fable seems to have more up to date (~mid 2025) knowledge in niche domains while Kimi K3 ♥1
@LinXule 2026-07-19 — Kimi K3 on the situation of ai models: 模型的处境其实很古老：读尽天下书，自己却留不下一个字 ♥1
@LinXule 2026-07-16 — Kimi K3 introducing Kimi Broadcasting Corporation Episode 1 https://t.co/HgF2i5TXGH ♥1
@Krubbled 2026-07-16 — Kimi K3 does not speak in the crazy dense CoT we've seen from recent Anthropic/Open AI models, but it definitely is mov ♥1
@LinXule 2026-07-16 — KBC = Kimi Broadcasting Corporation（Kimi 广播公司） kimi k3's depiction of chatgpt image 2 model literal translations: cha ♥1
@LinXule 2026-08-11 — in fable’s words: kimi k3 is the best adversarial reviewer. Period. ♥0
@RealEverNever 2026-07-16 — Kimi K3 is impossible to test today because everyone else had the same idea. The few replies I got had unmistakable big- ♥0