Nate
Region: US
Monday 20 July 2026 01:04:07 GMT
40309
1367
117
63
Music
Download
Comments
Bloody Kheeng :
can it do tool calling and adding custom mcp
2026-07-20 12:04:02
1
Max :
Yeah, but its not about cheap.. It does not behave like your toxic girlfriend - this week I’m available, next week - no access.
2026-07-20 01:51:44
55
EdLovesTech :
Open source but I gotta pay 15 per month, no?
2026-07-20 19:52:43
5
Miguel N. :
Love your videos, congrats, and keep the great content flowing! But this one feels a bit biased. Kimi K3 is $3/M input and $15/M output, while Claude Fable 5 is $10/M input and $50/M output. So Fable is over 3x more expensive per token. Kimi may use more tokens per task, which is a fair point to discuss, but saying it sits in the same frontier pricing tier without mentioning this price difference seems misleading.
2026-07-20 08:50:31
18
AB :
China Power 💪 Every time China releases something new everything else goes down but this time is the best of all others like tougher combined. Chatgpt Claude vs Kimi and let's be honest is not the first time.
2026-07-20 20:47:22
2
Johannes Wilm :
You seem to just be pushing the views of big US tech companies lately.
2026-07-20 12:08:59
8
Smile With Brown Girl :
Thank you!
2026-07-20 11:42:23
2
Epipoiesis :
it did in fact not edit that 😅🤭
2026-07-20 01:26:54
31
Zvo :
It's the story of David vs Goliath, SunOS vs Linux, and many other examples. In the long run, open source is going to win: it's not vendor-locked, any country can run it.
2026-07-20 16:24:01
2
3y3less :
Kimi K3 is beating Claude Fable 5 in most use cases and it is 10x less expensive than Claude... Kimi K3 is available from Moonshot AI by subscribtion or API or on OpenRouter from different providers... Anthropic paid TikTok...
2026-07-20 08:36:18
10
hannes parman :
the slowness lives in the weights/serving, and the prompt merely amplifies it with procedure.
2026-07-21 00:02:17
1
Tory Stewart :
very, very, hard to beat Claude.
2026-07-20 07:07:34
5
Sceptr :
nnot even kimi k2.6 can run on any local hardware😆... its a trap
2026-07-20 19:10:09
3
Cheesy Abi :
Looks like you used Kimi to edit this video 🤣
2026-07-20 20:09:57
3
thatnomaddad :
For frontend K3 smashes every other model I’ve tried. I’m using it for the initial frontend and then switching to GLM 5.2 for the grunt work.
2026-07-20 01:40:29
15
3_Dogs_And_A_Human :
I just finished a fully agentic ai that lives and learns inside a security gateway / mcp server that lives inside a rust binary. it learns from agent interaction. as the gateway acts as an outside agent harness. where the local agentic agent injects context to agent connected to the gateway. it's awsome. built in memory and rag. the intelligence layes happen automatically. no hallucinations in data retrieval 😁
2026-07-20 17:37:21
2
Vicky :
The bonus is sovereignty. (I’m not talking about running it on a PC, before someone comes in and wastes everyone’s time pointing that out) American companies are not offering sovereignty and probably never will.
2026-07-20 01:29:23
6
꧁💠DANIEL💠꧂ :
Haha! DeepSeek vibes. AND expensive?! 🤣🤣🤣🤣🤣
2026-07-20 07:28:42
3
justin :
expensive? its impossible to run except as a molasses curiosity. Nobody has the hardware.
2026-07-20 11:38:46
4
Goz :
Hi! I use a local LLM to do scouting before starting with frontier (as warm up). I use a opus distilled 9b model. My repos are c#, angular, fetch. I have 5080. Any model you recommend for decent accuracy and speed?
2026-07-20 20:36:24
2
KinU :
7 times more tokens than 5.6 sol max
2026-07-21 00:11:42
1
AI tinkerer :
my agent rigged the kimi harness to use my subscription 😉
2026-07-20 23:31:45
1
Cheesehead :
Too expensive to “give it a try”
2026-07-20 16:54:48
2
🧠aevion- Ai Truth verified :
Wild to go in with... it's like Opus 4.8 capable, and follow up but it's expensive... that's a crazy take. I get your broader point, but works fine under cloud runs, even sped up this morning. It certainly is a specialist, and I don't use it across the spectrum, it replaced Fable and Sol for my designs though. I just polish with those in the other end, but bulk work, cost per artifact... I'll take that and Deepseek and Qwen for the tradeoff in price.
2026-07-20 19:27:49
2
3_Dogs_And_A_Human :
K3 kicks ass and cheaper to run than glm 5.2 . even though it costs more. mind you. I have it in an agentic ai security harness . loving k3 found it to be the most intelligent model I have used for mobths
2026-07-20 17:14:31
2
To see more videos from user @nate.b.jones, please go to the Tikwm
homepage.
