TikTok | Learn | Advertise | Unsubscribe
Hi {{First Name|Guys!}}!
New leaders this week in AI:
China’s Kimi K3 - World 1st open weight model beats Fable 5 & GPT 5.6 Sol in frontend coding
Thinking Lab releases AI you can download & modify
Why I’m loving Hermes Agent for 1 person business
China’s Moonshot Kimi K3 (Why the Noise?)
China’s Moonshot AI just dropped Kimi K3, which is an important inflection point for AI.
It’s a 2.8 trillion-parameter AI model that is being positioned as one of the strongest open-weight models ever released.
Early benchmarks show Kimi K3 beats Fable 5 and GPT 5.6 Sol in front end coding! That’s a 17 place jump from Kimi K2.6 just 6 weeks ago.
While frontier level AI models still dominate majority of benchmarks, Kimi-K3 was able to rank #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Simulations, and Content Creation Tools, landing #2 only in Gaming behind Fable 5.
Based on Moonshot’s internal benchmarks, it’s not claiming it is the best but does seem to do well in 3D reasoning, coding, and vision.
Why it’s a big deal:
Chinese AI labs race is rapidly making US frontier level intelligence cheaper and more widely available. The gap is closing in fast.
This release sparked major sell off in global semiconductor and AI stock market (drawing parallels from earlier “DeekSeek moment”)
What people have built
Autonomous chip design in 48hrs - Designed a real functional chip
Excels in 3D modelling and animations (see below)
Some are questioning why they are paying double the price with Fable 5 with similiar outputs from Kimi K3
The Catch:
Not as cheap: While it’s cheaper than Claude’s Fable, it’s not as cheap for a Chinese open weight model. It’s priced similiar to what Claude’s Sonnet 5 will charge soon ($3/ $15) yet early testers say it feels like a Fable/Sol tier level
K3 is slow: Thinking effort is locked to max and you can’t lower it - but if output is great on first try I would argue slow beats cheap
Where to use it:
Kimi.com and Kimi App
Open weights on Hugging Face by July 27
This Week’s Scoop
Thinking Machines Lab (by former OpenAI CTO, Mira Murati) drops its 1st open model, Inkling, taking a different approach to OpenAI & Anthropic where companies can download & modify it directly.
Interesting to see it also partly used Moonshot AI’s Kimi K2.5 in early training proving China’s open weight models (ie. Deepseek, Alibaba, Z.ai, MiniMax etc) can help everyone progress AI developments faster.Grok Build uploaded entire codebases secretly
Gemini 3.5 Pro launch delayed due to fixing coding ability
OpenAI launches hardware for Codex - Keyboard with “yeet” key with smart speaker to chat with ChatGPT coming out next year
9 Reasons Why I’m Loving Hermes
If you’ve tried OpenClaw and gave up on it, it may be worth giving Hermes a shot. To be honest OpenClaw felt like magic but set up took forever and security measures were a nightmare. Not to mention it got progressively worse with gateway constantly breaking.
Hermes is what OpenClaw should have been so yes this does feel like a personal AI agent BUT it learns from you over time. My video here.
Here’s why I’m loving Hermes as a solopreneur after testing:
Super easy set up - Luckily now you can get Codex or Claude Code to go and set it up for you 🙂
Security - Better than OpenClaw which was a nightmare. While it is safer in general, you will still want to set this up on either mac mini/old laptop or cheap VPS ($5/mo).
Give it a separate email, phone number and credit card. There’s AgentMail for emails, AgentPhone for phone numbers for AI agents. I just found out about it so I haven’t used it but for me I just gave it a free gmail account and bought a cheap $2 simcard to start. Not necessary to start but incase you were wondering.Desktop App - Yes! You don’t need to use the terminal and it feels like Apple/OpenAI “esk”
All in one place - In the desktop app you can see all your skills, cron jobs, chats, chats with telegram (Slack, iMessage etc), artifacts (all your images, files, docs)
Skills/ tools already inbuilt like computer use, browsing, automation, create images/video, transcribe
Persistent Memory - Constantly learning and self improving + building you skills without you needing to tell it. We all love that coworker that just gets it
Gateway not as shakey - Desn’t break as much as Open Claw
Able to chat to it literally remotely without having to remote access every chat like Claude Code. Codex is better but both are still fiddly and doesn’t work the way it should. Hermes actually works
Kanban board - If you’re visual like me, you will love the ability to tell it to add tasks to the board so you can see what it’s working on, progress, completed and each box tells you details!
If you wanted to play around with it over the weekend, here’s where you can access Hermes.
That’s all for this week!😊
Cheers,
Nat
Did you Enjoy this Week's Newsletter?
🚀Ways I Can Help You
Book Newsletter Sponsorship (57,000+ subscribers, 37% open rate)
Book Brand Sponsorship (over 900,000+ followers on IG/Tiktok/FB/Linkedin)
