Ambassador @DevinAI @cognition building @Sen_computer
Germany
Joined October 2024
- Tweets4.5K
- Following418
- Followers1.2K
- Likes7.8K
Pinned Tweet
CLAUDE'S 5-HOUR LIMIT IS BACK TO NORMAL 🔥
A few days ago I said the session window felt nerfed.
Now it's back.
Even the $20 plan gives you solid Opus 5.5 usage again.
Anthropic listened, or fixed it quietly.
GPT-6.1 TERRA is CRAZY 🔥
OpenAI cooked with this one.
For computer use it's honestly incredible.
Fast, focused, and it just gets things done.
But let's be real about the new Artificial Analysis Index v5.
It ranks Terra (51) above Fable 5.1 (49).
That doesn't match reality.
Anyone who has used both knows Fable is on another level.
Benchmarks aren't everything.
But for computer use, Terra is a beast
GOOGLE JUST DROPPED GENIE 4 AND IT LOOKS UNREAL 🌐
This isn't a video you watch.
It's a world you walk through.
Live and in real time.
The scene reacts to you.
It has sound.
People in it move and react.
Honestly, I can't wait to test it.
Fingers crossed EU users get access too.
And if this keeps improving, does Google even need separate video models anymore?
Maybe one day it's all just Genie.
Well done google
Artificial Analysis Index V5 is coming today.
Let’s see who paid the most to rank higher.
Just joking.
WHAT IF YOU NEVER HAD TO SWITCH APPS AGAIN? 🤯
Your personal assistant.
Your coding agent.
Your custom bots.
All in one single app.
Connected to every device and system you use.
Running Codex, Claude Code or Devin, right inside.
Sounds good?
Stay tuned.
New openAI models feel like this:
> GPT-6.1 Sol = old Terra
> GPT-6.1 Terra = old Luna
maybe this explains the "price reduction"
GPT-6.1 Terra is really fast and amazing at computer use.
Unpopular opinion: You probably don't need Claude Fable 5.5.
Everyone is refreshing the timeline waiting for the release this week, and I get the hype. Sure, it’s going to be a monster model.
But genuine question for builders: What are you actually trying to build right now that Opus 5.5 can’t handle?
Opus 5.5 is fast, way cheaper to run, and already good enough for almost every real-world product and workflow out there.
Be honest: what is the one thing you're blocked from building today because Opus 5.5 isn't smart enough?👇
DEVIN DESKTOP JUST GOT A MASSIVE GLOW-UP 🔥
OGs still know it as Windsurf.
Now it finally has a brand new UI, and the UX is so much better.
It just feels smooth.
Most people don't know this:
Devin Desktop supports almost every ACP agent out there.
Claude Agent, Codex, OpenCode and many more.
Plus the best connection to Devin Cloud, by far the best coding agent in the world.
Give it a try.
SWE-2 is still free on paid tiers until the end of the month.
KIMI K3.1 PRO IS OUT 🔥
This morning it was a teaser video.
Now it's live in Kimi Code AND API!
K3.1 Pro
K3.1 Pro Highspeed
Low / High / Max thinking
So far it feels really good.
Token efficiency is actually nice.
It is not sitting there overthinking forever.
It just gets to work.
That is the upgrade I wanted.
Not another model that burns 2 minutes to write 3 lines of code.
Moonshot blog post and benchmarks should come soon
It has heavy GPT-6 Astra feeling.
KIMI K3.1 IS ABOUT TO DROP 👀
Moonshot just posted a silent teaser.
Terminal says: Model: Kimi K3.1
Arena already spotted it:
"A new model from Kimi is coming."
People are saying it will be way more token efficient and cheaper.
Honestly that is what I want.
Not another model that thinks for two minutes to write three lines.
Overthinking models feel smart until you wait forever for the answer.
Also watching:
Qwen 4
GLM 5.5 👀
MiniMax M3.1
China cooooooks
OPENAI JUST MOVED INTO GOOGLE'S HOUSE 🤯
OpenAI announced today that its frontier models and Codex are coming to Google Cloud.
And it goes further than that.
OpenAI models will soon run inside Google Antigravity and the Gemini Agent.
Yes, Gemini's own agent will be able to use GPT.
Google Cloud CEO Thomas Kurian:
"so they can choose the right model for any given job"
The model war just turned into a distribution war.
Companies don't want to switch clouds for a better model.
So the models are coming to them.
Compute is power
Justin retweeted
astra went from "greatest model ever made" to "biggest disappointment of the year" in 6 weeks.
I love AI twitter.
Justin retweeted
sent from my ipad
(minecraft java edition ported to rust ported to iOS, running natively with no JVM/JIT)
.@strellic and I did a full 1:1 file-by-file port of Minecraft from Java to Rust
both the server and client are ported -- there are some perf bugs but it does appear to work!
EVERYONE WANTS FABLE 5.5. I WANT A CHEAPER OPUS 5.5 💸
Hot take:
If Fable still only gets 50% usage, it can be the smartest model ever made.
We just won't get to use it much.
Opus 5.5 is already incredible.
Most people don't need a smarter model.
They need one they can actually use all day.
The next big leap isn't more intelligence.
It's lower prices.
ASML JUST SAID THE AI CHIP SHORTAGE LASTS UNTIL AT LEAST 2028 😳
Every answer from ChatGPT, Claude and Gemini runs on chips that need one company's machines.
That company is ASML, the only one on earth that makes EUV machines.
Without them there are no Nvidia GPUs, no Google TPUs and no HBM memory.
In July they planned +30% more machines for 2027, from about 65 a year to about 85.
On Oct 14 CEO Christophe Fouquet said:
"our supply as currently planned for both 2027 and 2028 will not fully meet the demand of our customers"
Customers already want to place orders for 2029 and 2030.
TSMC, one day later: "the gap between demand and supply are still quite big"
Micron: HBM for 2027 is largely sold out, at much higher prices.
My take:
A new machine takes 1–2 years to turn into chips, so the extra capacity really lands in 2028–29.
For ChatGPT and Claude users, don't expect rate limits or prices to get much better in 2027.
For OpenAI, Anthropic and Google, whoever locked in chips early keeps the lead.
Smarter models will come from better training, not just more chips.
But anyone hoping for cheap, unlimited compute in 2027 can forget it.
Sources:
ASML Q3 2026 results (Oct 14):
sec.gov/Archives/edgar/data/…
Reuters analysis on ASML's sold-out outlook:
investing.com/news/stock-mar…
TSMC Q3 2026 call transcript:
investor.tsmc.com/english/en…
Micron on HBM sold out for 2027:
arstechnica.com/information-…