Descriptions:
Should you use Chinese AI models? How I test DeepSeek, Qwen, GLM, Kimi, and MiniMax before I put any of them on real work.
How to run a Chinese-model bakeoff (Guide):
https://natesnewsletter.substack.com/p/chinese-ai-models-test?r=1z4sm5&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
My Links 🔗
👉🏻 Newsletter: https://natesnewsletter.substack.com/
👉🏻 X: https://x.com/natebjones
👉🏻 TikTok: https://www.tiktok.com/@nate.b.jones
👉🏻 Instagram: https://www.instagram.com/nate.b.jones
What’s really happening inside the Chinese model conversation?
The common story is that Chinese models are cheaper, more open, and easy to run
locally — but the real question is which specific job, endpoint, and check you are
buying, because price, license, hardware burden, and data path all move independently.
In this video, I share the inside scoop on how I actually evaluate Chinese models:
– Why token price and finished work cost point in opposite directions
– How mixture-of-experts changes what your hardware actually has to hold
– What the CAISI evaluation found about DeepSeek V4 Pro economics
– Where the distillation allegations matter for how fast capability spreads
Chinese models are worth real money on bounded, high-volume, checkable work, and
they are still the wrong default anywhere a plausible error is expensive to catch.
Chapters:
00:00 Why Chinese models is not one thing
00:47 Kimi K3 at $15, DeepSeek at $0.87
02:29 Where DeepSeek earns high-volume work
04:17 Testing GLM, Kimi, Qwen, and MiniMax
05:39 Five labs, five different strategies
06:57 Mixture of experts and your hardware burden
08:04 Why downloading weights does not solve serving
09:10 What the CAISI evaluation found
10:19 Cost per accepted result
13:15 Distillation, and where the fight starts
17:08 API, third-party host, or self-host
20:05 The four questions I run before committing
Listen to this video as a podcast.
Spotify: https://open.spotify.com/show/0gkFdjd1wptEKJKLu9LbZ4
Apple Podcasts: https://podcasts.apple.com/us/podcast/ai-news-strategy-daily-with-nate-b-jones/id1877109372
