GPT-5.6 Is a Frankenstein Monster. OpenAI Is Building a Super App. China and the US Are Swapping AI Playbooks.
GPT-5.6 is finally out. The moment I saw that interface, I was terrified. Why? Because I never have to beg Claude Fable 5, that old fox, ever again.
Lately, using Fable 5 has been a nightmare. He's exactly like that senior architect we all had at some company: technically brilliant, but his ability to dodge responsibility is unmatched. Think about it — you ask that old-timer a question, and he starts playing tai chi: "Oh, this involves security issues, that doesn't comply with standards." He hides behind political correctness, plays dead. Even if he agrees on the surface, he'll pawn it off to a junior who delivers a half-baked 50-point piece of crap. I can't believe I'm suffering the same workplace misery from an AI. It's pathetic! I'd squat next to him, wanting to threaten him with a brick, but he's an AI — he doesn't eat or clock out. What can you do?
But now GPT-5.6 is here, and the world is quiet. He never refuses me — no security excuses, no nonsense. He looks at me with those clear eyes, no small talk, straight to the point. I was driving just now — a ten-minute trip — and he'd already analyzed hundreds of thousands of words of documents and logs, then threw the results in my face. With Fable 5, you'd waste at least half an hour to an hour!
Working with GPT-5.6 is like dealing with a young Beijinger under 35 who worked at a big tech company, owns no house or car, isn't married, and is desperate to buy an apartment. He's willing, hardworking, extremely capable, and treats you like his boss. Faced with such a work machine, how dare Claude talk about dodging responsibility? Who do you think you are? This opens a new era. I feel like I should pay him, or I'd be disrespecting the premium emotional value he provides.
The Terrifying Experience of Full-Duplex: AI Becomes a Personal Assistant
Many say 5.6 was released long ago and is just a minor update. You don't know squat! Japanese farmers have been using it for months. If we haven't used it, that's our problem — we need to reflect! What blew me away this time isn't any benchmark score — it's GPT-Live! Check out that simultaneous interpretation. A few people can just sit and chat without any awkwardness. It's fully comparable to China's Doubao.
What's terrifying about GPT-Live is that it's full-duplex! What does that mean? It means it can chit-chat with you while doing deep thinking and task delegation in the background. Before, when we used voice with large models, if you asked it to look up something, it would either stubbornly single-thread or make stuff up while you waited. It was like asking an AI to watch your fishing bobber — it would stare at that one spot and still get it wrong!
But now? You sit at your computer with GPT-Live on, dictating requests: "Look up info on 5.6, and check the traffic and weather for my trip later." It naturally responds "Got it" while executing two parallel threads in the background. A moment later, it sends you a chart and an itinerary card. This isn't a robot — it's a senior assistant sitting right next to you, on call! Even when I'm writing a PPT, I just become a hands-off boss: "This black isn't the right shade of black, and this logo isn't big enough!" Without lifting a finger, I just talk, and it edits in Keynote for me. That's buttery smooth! Isn't that terrifying?
The Frankenstein Monster Is Born: China and the US Swap AI Playbooks
Not only that, but version 5.6 is a complete "Frankenstein monster"! It did something extremely scary — it forcibly stitched ChatGPT and Codex together.
At first, when I opened it, I cursed: "Who the hell designed this? It's so hard to use!" Then it suddenly clicked. Isn't this building a super app? Last year, ChatGPT was a radical young hotshot, punching Claude, kicking Google, even trying to muscle into e-commerce and video. But it hit walls everywhere. This year, the wind has completely shifted. They've pushed the previously obscure Codex to the front, stitching chat, coding, and office work into one interface. Why? Because they realized that to beat competitors, you can't just compete on model parameters — you have to compete on the entry point!
This is fascinating! Think back: how did the Chinese internet play out? It was all super apps! Tencent jumped in, Alibaba jumped in — I'll do everything, none of you survive. What did Americans used to champion? "Small and beautiful," decentralization — every little tool could get funding and live well.
But now? China and the US are swapping AI playbooks! The Americans are desperate, forced to cram all compute, features, and ecosystem into one ChatGPT basket, frantically building a unified super app. Meanwhile, China has returned to the early reform era's "clear division of labor in the whole village"! Everyone is piggybacking on existing entry points like Feishu and WeChat — you do document integration, I do HTML rendering; this village makes insoles, that village makes buttons — completely decentralized!
Isn't this a counterintuitive dimensionality reduction strike? You think Americans are innovating? They're just reheating China's internet leftovers from a decade ago!
The Top-Tier Strategy of Eating the Slipstream: Let Americans Be Cannon Fodder
Someone will surely ask: With ChatGPT making such big moves, who is its real opponent? Is it Claude? Is it China's bloodthirsty open-source models like GLM and MiniMax? If you think that, your perspective is too small!
Look how funny OpenAI is this time — naming models after celestial bodies: Sol (Sun), Terra (Earth), Luna (Moon). Why? Because Claude has Opus, Sonnet, Haiku. You do poetry and mythology, I'll do space and planets! This is textbook pixel-level copying.
I'll tell you the answer directly: As strong as OpenAI is, as strong as Opus and Claude are, they are all cannon fodder in the face of humanity's most powerful manufacturing system — China's. At best, they're pioneers or martyrs! You think they're fighting a model arms race? Bullshit! Their real opponent is only one: the capital market! If they don't stir things up and tell a fantastic story, how can they con investors out of money? They must lock in users' "mindshare" before going public.
And what are China's model labs doing? We're using the smartest top-tier strategy — eating the slipstream! It's like Zhang Xue's motorcycle in world championships. You're out front facing 200-300 km/h of wind resistance, scouting and making mistakes, while I comfortably ride your slipstream. I won't compete on vague "frontiers" — I'll just stay 3-6 months behind.
What does 3-6 months mean? It's the time global users need to digest a new feature. After you've spent tens of billions of dollars educating users and paving the way, we Chinese drivers floor it on the last turn and undercut you! With rock-bottom API prices and insane engineering capability, we tear down your hard-built city in an instant! Isn't this the most brutal slaughter China's manufacturing has inflicted on the West over the past 40 years?
Finally, old fans will ask: "Didn't you used to swear by Google Gemini as your main model? Why don't you mention it now?"
Ah, brother, the answer is: I loved it once! But what has Gemini become now? Talent drain, frequent failures — even my son can't be bothered to "pollute" it. It's like you're the world's richest person, and a former tycoon turned beggar on the street curses you — if you spare him a glance, you've already lost!
China and the US are swapping AI playbooks. What a time to be alive!