Kimi K3 shipped and a competitor's stock lost 30% the same week. It wasn't an American one.
Z ai fell as much as 30%. MiniMax 16%. Alibaba 4%. The US labs it was supposedly aimed at aren't even listed.
K3 is 2.8 trillion parameters, the largest open-weight model ever released. Roughly 75% bigger than DeepSeek's V4 Pro.
One million tokens of context, native vision, thinking mode always on. #1 in Frontend Code Arena, top three on every major index.
Then Moonshot gave the weights away on July 27. Free to download, free to self-host, yours forever.
That reads like generosity until you check who it actually hurt.
To a frontier lab, an open model is competition. To the second-best open model, it's an extinction event.
Nobody pays for mid-tier open weights when better ones cost nothing.
The license is where the business hides. Past $20M in revenue you negotiate a contract. Past 100M users you display their name.
Free for you. A contract for anyone big enough to matter. Daily revenue is up at least 6x since launch, and they're raising at $50 billion.
Open weights stopped being a gift. They're a pricing weapon now. Go look at what your stack runs on and ask what the giver gets back.
an agent is four parts in a loop. you own one. the other three break it.
that's why the demo works and prod doesn't. you can't debug what you can't see.
1) the prompt → what you tell the model each turn. you own this one. good.
2) the context window → what it sees right now. the framework fills it with junk, and you never notice until it rots.
3) the tools → what it can do. you own the list, not when or why it fires them.
4) the control flow → what happens next, when to stop. the framework owns this. it's what breaks at 80%.
own all four and your agent stops being a magic trick that works on stage and dies on call.
this isn't my idea. it's the 12-factor agents guide (24k stars)
github: https://t.co/bvj2cxGjgC
the whole thing every serious builder ends up rewriting their stack around.
full breakdown in the article below.
China open-sourced a model that reconstructs any scene in 3D from a regular video, in real-time.
one camera. no LiDAR. 10,000+ frames without falling apart.
just walk around with your camera and watch the entire world get rebuilt in 3D at 20 fps.
→ runs at ~20 FPS on a single GPU
→ Stable over 10,000+ frames
→ Beats optimization-based methods on benchmarks
→ Works on drone footage, driving videos, indoor walkthroughs
100% open source.
This 25-minute Claude Code workshop by Anthropic's own applied AI team will teach
you more about Claude Code best practices and making your AI tools actually work together than everything you've scrolled past this year.
Bookmark this & watch, no matter what.
Then read the guide below.
People think learning AI takes months.
It's really just a couple of hours.
And I wrote 17 free guides to start right away:
Claude 101: https://t.co/Jv1jsvFB7T
Claude Code: https://t.co/WYZd5ltnXo
Claude Skills: https://t.co/jT4uB5AFtY
Nano banana 2: https://t.co/qfHT594CCI
Claude in Excel: https://t.co/mfcXYSA57j
Best AI for Search: https://t.co/77BmjbJjP0
1M followers with AI: https://t.co/1TV9LYAptv
Claude for your team: https://t.co/U1JsBVC299
No prompt saves you: https://t.co/SDKJWykHE4
AI Slides (PPT in 2026): https://t.co/RfcyYRQ2Ad
Set up Claude Cowork: https://t.co/diDhiKjHtU
Claude to sound like you: https://t.co/99RzxXTvzs
Claude interactive charts: https://t.co/ebCHGZqgPt
Claude as your computer: https://t.co/TxYuHPiImn
Claude Cowork + Project: https://t.co/Q7AN9CZ2mg
You're an AI workaholic: https://t.co/mCIvB3ZPA5
Setup AI before prompting: https://t.co/pE3OF722aw
___
1. Save this list for later (three dots, top right).
2. Share it with a friend by ♻️ reposting this image.
3. Subscribe to my free newsletter: https://t.co/psB7XxAv8w.
AI ending interior design
Nano banana 2 now can turn sketch floor plan into 4K 3D rendering with accurate dimension, take photos for each room, and 1-click furniture change
used to cost $100k and months.. now cents and mins
step by step tutorial on OpenArt:
This army of @openclaw agents runs an entire company for $400/month. Here's the exact structure to follow.
(bookmark for later)
1/ Core
→ Jarvis (the brain)
→ Model: Opus 4.6 via Claude Max OAuth
→ Routes every task to the right sub agent automatically. YouTube URL comes in, it goes to Clipper. Research report lands, it goes to Scribe. All task routing logic lives in structured MD files the agent reads from.
2/ Research
→ Atlas (deep research analyst)
→ Model: Claude via OAuth
→ APIs: Brave Search, X API, FireCrawl
→ Cron: Every 1 hour
→ Runs deep research across X, Reddit, and the web nonstop. Trained on
MrBeast's virality framework from every podcast he did on YouTube analytics, plus Dan Koe's viral article structure. Outputs research reports and a master virality playbook MD file that the content team pulls from.
3/ Content
→ Scribe (copywriter)
→ Model: GLM 5
→ Cron: Every 3 hours
→ Takes research from Atlas and writes draft posts matched to the founder's voice and style.
→ Trendy (trend scout)
→ Model: GLM 4.7
→ APIs: X API
→ Cron: Every 2 hours
→ Scans X and Reddit for trending topics and viral patterns. Reports findings back so Scribe can write timely content around what's working right now.
4/ Design
→ Image Designer
→ Model: Nano Banana Pro (Google API)
→ Generates images on demand.
→ Video Producer
→ Models: Higgs Field API + Brok Imagine API
→ Creates AI UGC videos and video content.
→ Motion Designer
→ Model: Claude Code (OAuth) + Remotion
→ Produces motion graphics and animated content.
5/ Development
→ Clawed (senior developer)
→ Models: Claude Code (OAuth) + Codex 5.3 (API)
→ Cron: Every night at 11pm
→ Reviews entire codebase, identifies what's missing, and ships pull requests by morning. First feature it ever built was a FAQ section it realized the homepage needed. Spins up multi agents within Claude Code so one reviews, one builds, one handles security in parallel.
→ Sentinel (code reviewer + bug monitor)
→ Model: Separate LLM (acts as second review layer)
→ Cron: Every 2 hours
→ Reviews all pull requests from Clawed before anything gets merged to GitHub. Also monitors production for user reported bugs and errors.
6/ Growth
→ Atlas + Scribe working together
→ Atlas finds Reddit threads where people complain about competitors or ask for clipping tool recommendations. Scribe drafts responses. The founder copies and posts. This workflow alone drove 450+ users to the SaaS with zero ad spend.
7/ Operations
→ Clipper (clipping agent)
→ APIs: Poster API
→ On demand (triggered by Jarvis when a YouTube URL is pasted)
→ Takes YouTube URLs, clips them, adds captions, and auto schedules posts to social channels.
→ Ryder (9 to 5 support)
→ On demand
→ Handles tasks for the founder's day job. Article writing, research, daily work support.
The breakdown: 6 agents run on Claude models.
The rest run on cheaper API credits across GLM, Higgs Field, Brok Imagine, and others.
This is how solo founders are running entire companies now. The team is already built. You just have to set it up.