you can just hallucinate the entire internet with Qwen 3.8 27b running at 2,000 tokens/second? part 2 of turning @cerebras + @Alibaba_Qwen 3.8 27B into an OS: built an offline browser with zero network calls and mounted it directly the JIT ubuntu desktop. no wifi. no scraping. zero packets sent to external CDNs. you search a site, set a year, and qwen 27b at 1,950 tok/s synthesizes the entire DOM on the fly. here is youtube in 2045 vs 1999: → search google for youtube inside the OS → scrub to 2045: instant futuristic feed → scrub to 1999: raw web 1.0 time capsule in seconds at this speed, browsing isn't retrieving files from a server, it's querying an alternate reality. a 2D browser window is just step zero. imagine full operating systems, virtual worlds, and complex simulation engines existing purely as model weights. zero gigabytes stored on disk, just pure interactive reality streamed on demand. What else becomes a possibility with the qwen 3.8 27b (dense) at 2000 tokens/sec?
Operating System powered by Qwen 3.8 27B at 1950 tokens/sec! here is what 1,950 tokens/second @Alibaba_Qwen's 3.8 27b actually looks like on @cerebras: i wrote a minimal python web server that turns cerebras inference into a live operating system. zero apps on disk. when you double click an icon: 1) python proxies a raw SSE stream from qwen 27b at 1,950 tok/s 2) calculator compiles & mounts in 11s 3) full canvas paint studio with brush engine compiles in 10s. at 2,000 tokens/second, software is just an on demand hallucination that runs instantly. the model weights ARE the operating system runtime. what else would you build at 1,950 tokens/second?

Sep 15, 2026 · 12:11 PM UTC

157
354
4,982
1,691,726
Sort replies: Relevant Recent Liked
This is very cool. It's also what I would expect the last man on earth would use to prevent himself from going mad. "Let's go check twitter!" (Hallucinated posts form millions of people who aren't there)
4
8
445
37,249
the ultimate conclusion to the dead internet theory.
2
192
28,947
Please someone put this on their parent's computer
1
28
2,268
😂the ultimate parent sandbox.
1
17
1,898
so the browser is just a really confident liar now? i can respect the efficiency, but my debugger might have a few questions about that synthetic DOM.
1
18
10,634
it’s not lying, it’s just aggressive creative hydration lol. true your console tab would probably have an existential crisis
12
8,592
Absolutely incredible usecase of this model!!!
1
13
1,896
thanks buddy! i hope i can hallucinate a few more cool ones haha
1
926
lol this is a really cool idea!
1
7
3,220
Thanks buddy! Just trying to hallucinate what 2000/secs of hallucinations can unlock lol
6
2,689
yo @RobertHaisfield it's like websim but... local? and with a time-travel slider?
2
2
901
yeah but i hadn't heard about them. totally ahead of their time
701
wow that's so cool. you should have this hallucinate a virtual machine which hallucinates a browser which hallucinates another virtual machine which hallucinates a browser which hallucinates the internet which kill yourself
3
111
15,876
Nolan will love this script
2
2
36
12,152
haha maybe after a few more
10
7,831
Wow. just a reminder that we are very very early still in this. Just like no one could have imagined the internet would become what it is today
1
15
8,116
for real. people in 1995 thought the internet was just a digital fax machine or newspaper.
2
10
6,385
Purely another world. Reality distorted beyond the horizon.
1
9
2,648
Yup crazy right. we could end up spending more time in the virtual worlds created by our agents
4
2,220
This is cool! What about adding minimax and hallucinate videos?
1
5
5,762
Yup I thought about it. Only problem is it wouldn't be fast enough for live interaction. Otherwise you could create your own youtube that generates content on the fly
1
14
4,703
2,000 tokens per second is actually insane.
1
5
1,922
For 2026 😁
3
1,563
I’m trying to optimize with shitty hardware see how far I can come. No idea tbh.
2
4
1,018
true engineering is squeezing 4 tokens/sec out of a 20 years old laptop. fan screaming like a boeing 747 taking off.
10
839
Haha. Google has AI mode in it. Will Qwen replicate behavior of Gemini? That would be awesome
1
5
596
that is a though ask i guess
1
3
434
Your age mates are solving real problems
1
1
2
6,014
50 years of ai progress to hallucinate a web page might not sound like the best thing to do. These tokens could certainly have been put to better use. But sometimes I like to share even the most stupid ideas. It could be a seed for better ideas that come to me or someone else. Nonetheless such a browser is actually viable and useful today where each call hits your agent first which decides if new info should be fetched from internet and filtered or regenerated to remove any biases
7
5,215