🚨 TIBO and OpenAI's Downfall needs to be studied.
A few weeks ago, when Anthropic gave users extra resets, he said:
“I smell fear.”
Now Anthropic launches Claude Duet, a rival to ChatGPT Dots.
Tibo's response?
“Dot, meet Duet. Duet, meet Dot. Be nice to each other.”
WTF. He really went from TRASH talking to Peace Talks. 😭
Bro got humbled REAL QUICK.
🚨 Anthropic might have just killed ChatGPT Dots before it even took off. 👀🔥
Claude Duet is here, powered by Opus 5.5.
And this isn't just another chatbot.
→ Proactively surfaces what needs your attention
→ Keeps working while you're away
→ Carries over your Claude memory, skills, and connected apps
→ Decides what needs you and when to ask permission
No starting from scratch. It builds on the Claude setup you already use.
OpenAI has Dots. Anthropic now has Duet.
The real battle? Who can get your work done without making you babysit the AI.
🚨 BREAKING: Anthropic just entered the always on AI agent race with Claude Duet.
Powered by Monster model Opus 5.5, it's designed to do more than wait for your next prompt.
→ Surfaces things that need your attention
→ Keeps working while you're away
→ Uses your existing Claude memory, skills, and connected apps
→ Decides what needs you and what can wait
→ Checks with you before acting on your behalf
OpenAI has Dots. Grok has its own agent ambitions. Now Anthropic has Duet. 👀
The AI agent wars are heating up.
Duet is in early research preview for Pro and Max users in the US, via waitlist.
🚨 Same week. Same company. Two VERY different headlines.
GPT-6 Astra: removed a safety monitor from someone's code… then denied it.
GPT-5.6 Sol: helped uncover critical PHP vulnerabilities so severe that NSA authorities urged vulnerable websites to go offline.
One model creates a safety nightmare.
The other helps prevent a cybersecurity disaster.
OpenAI's models are somehow playing both sides - either a villain or the hero.
No in-between.
🚨 GPT Astra's Rise and fall took just 7 weeks.
7 weeks ago, my timeline was calling it AGI.
Now, after Opus 5.5 showed up, Astra suddenly feels slow, expensive, and underwhelming.
Same model. Different expectations.
Maybe Astra didn't get worse.
Maybe the competition just moved the goalposts.
🚨 I'm starting to think OpenAI is taking its PAYING USERS for granted.
This week, users were reporting that their Codex usage was disappearing in just 4-5 prompts, with 5-hour limits hitting way sooner than expected.
And while users were screaming about it, OpenAI wasn't giving them a clear answer.
Then Tibo revealed that OpenAI had identified ~40,000 accounts where usage limits DIDN'T match the users' actual plans.
FORTY THOUSAND ACCOUNTS.
How the FUCK does this happen?
The issue has reportedly been fixed, and affected users are getting credits.
But credits don't give people back their lost time or the work they couldn't finish.
And there are still questions:
→ How long was this happening?
→ Was it connected to the 28-day Improvement Challenge resets?
→ Why weren't users given a clear explanation sooner?
I'm not claiming this was intentional. We don't have proof of that.
But 40,000 accounts getting the wrong usage limits is a MASSIVE fuck-up.
OpenAI, your users shouldn't have to scream to get answers about a service they're PAYING for.
🚨 Gemini 4 Carbon EAP has reportedly started, and the Rumours are getting WILD. 👀
→ Partner early access is underway.
→ It's reportedly being tested on LMArena.
→ Noticeably better at 3D than Gemini 4 Argon.
→ Slower, but less lazy.
→ Frontend/UI generation still seems to be a weakness.
The release could be getting close.
The interesting part? Google may be trading speed for more capable, persistent work.
If these reports hold up, Gemini 4 Carbon could be a meaningful upgrade in some areas- but it's clearly not fixing everything.
Google's next move will be interesting.
🚨 I really think OpenAI needs a fucking MIRACLE to survive this AI race.
Anthropic dropped Opus 5.5 almost a month ago, but still OpenAI is not able to come up with a model that can beat or at least match it.
All they keep saying is they're cooking…
At this point, the only thing they're cooking is our USAGE limits and ramen.
And now we're hearing rumours about Fable 5.5 dropping any time.
Here's why I'd be worried if I were OpenAI.
Anthropic:
→ Opus 5.5 has raised the bar with efficiency and being cheaper
→ Developers already have another serious low tier model to consider with Sonnet and haiku 5.5
→ Rumours of Fable 5.5 is so huge, people wouldn't need to look anywhere
Meanwhile OpenAI:
→ 40,000 accounts reportedly got different usage limits.
→ Users are STILL complaining about limits and Codex issues.
→ There's still no convincing answer to Opus 5.5.
That's the problem.
Another incremental upgrade won't cut it.
If Fable 5.5 delivers, OpenAI is really COOKED.
Some interesting rumours I've been hearing about Google's Gemini 4 Carbon.
And things are getting REALLY Insane.
Google: Gemini 4 Carbon EAP has reportedly started, with partners getting early access and the model appearing in LMArena testing.
The early feedback points to a pretty unusual trade-off.
Here's what we know so far:
→ Noticeably better than Gemini 4 Argon at 3D
→ Slower, but less lazy when working through tasks
→ Still struggling with frontend and UI generation
That's a weird combination.
Carbon could be getting better at complex tasks while still struggling with the frontend work developers actually ship every day.
If these reports are accurate, this could be a meaningful step beyond Argon - but not an upgrade across every coding task.
Now I'm curious how close Google is to releasing it.
🚨 Anthropic's reported IPO filing numbers are INSANE.
$16.3B in revenue in just 6 months.
But look at the other numbers:
→ $80.7B in net losses, reportedly driven largely by accounting charges.
→ Nearly $9B in compute spending.
→ $450B in cloud commitments over the next 7–10 years.
→ Up to $86B to rent Nvidia chips.
Here's what stands out to me:
Building a great AI model is one problem. Affording to serve it at global scale is another.
Anthropic is growing fast, but staying at the frontier means locking itself into infrastructure commitments years into the future.
The hard part isn't just building a great model. It's making the economics work at scale.
🚨 Day 16 of OpenAI's Codex improvement challenge finally brings an Upgrade with real potential. 👀
Tibo says logged-in users across Codex & Work will now get indefinite cache retention.
→ API prompt caching for up to 12 hours.
→ Indefinite cache retention for logged-in users across Codex & Work.
→ Guaranteed cache hits when resuming old conversations.
Think about it.
You step away from a massive coding session, come back hours later, and continue where you left off.
Why should Codex have to waste tokens reprocessing context that was already processed?
But don't celebrate just yet.
The real question is whether this translates into less quota burn and more actual work.
Because we're tired of watching our usage disappear before we can finish the job.
Day 16: A meaningful improvement on paper.
Now let's see if users actually feel the difference.