Two days ago the US banned Claude Fable 5 because it was supposedly so smart that endangered the national security of USrael. Yesterday China released GLM 5.2. Today GLM 5.2 is #1 on @bridgebench. BS at 100.0, and #1 on Reasoning at 42.8, beating Fable 5. At 1/10th the cost and 300 tokens per second.The AI bubble burst and flushing entire Mutt-economy down the toilet cant happen fast enough xD
>>537067941>it was real in his headany luck on the toilet cleaning job in Liverpool?
sikhs are karmelo tier
>>537067941>I hope the mutt economy crashes and burns!You have a child's understanding of economics at best.
>>537068033>>537068041I stopped using shitty ChatGPT years ago. Out of Western (so de facto only American) models, i used only Grok, but not anymore as using it for free is now pain in the ass, as it constantly wants me to subscribe.I am using _exclusively_ Chinese models now. Get fucked Mutts xD
>>537068710>Chinese modelsAren't these just shoplifted US models...
>>537068840Yes, but the reverse is also true. They all steal from each other.The Chinese models are often significantly cheaper to run, they are often ahead in optimization.
American models are still supreme cope and seethe
>>537068600>You have a child's understanding of economics at best.Chinese chip independence + open AI models could cause/accelerate a US AI/compute bubble crash via cost collapse and market shift. US export controls pushed China to optimize algorithms and scale domestic chips (Huawei Ascend etc). Chinese firms achieved "good enough" performance with far lower inference costs via efficient architectures, mixture-of-experts and domestic silicon clusters. GLM 5.2 hits top BridgeBench scores at 300 tok/s and 1/10th price with 1M context and open weights outpacing restricted US frontier models in accessibility/speed despite potential gaps in peak training compute. AI valuation assumes massive GPU spend (Nvidia) for scarce, high-margin closed models. Chinese independence floods supply with cheap, open alternatives deployable on local/lower-end hardware. Developers switch for cost (1/10th) + speed, crashing margins on US cloud inference and hardware. Overhyped capex (data centers, chips) yields negative ROI as commoditization hits. Talent/investment flees to efficient/open ecosystems. US restrictions inadvertently sped Chinas self-reliance loop (data + optimized models on domestic chips), compressing timelines. This way "economy" of the Muttland dependent now on the AI bubble can easily get fucked in a very brutal way dwarfing 2008/2009 xD
ask it about tiananmen square protests or JUST TIANANMEN SQUARE SIMPLY AS A LOCATION and watch it clam up and block you / freeze / abort. open source? rly? huh.. fuck u
>>537067941The logic is simple, China will start publically stating they view AI as having genuine value in the battlefield, while claiming the west sees no value in its deployment
>>537069623still behind Claude Opus 4.8China loses again LOL
>>537069623This index doesnt prove US superiority because it aggregates diverse benchmarks (many non-reasoning, telecom, etc), not pure frontier intelligence. Chinese models like GLM-5.1 score competitively overall, many top spots use heavy US tech/infra. It measures accessible performance, not raw training compute.
>>537068840They all steal from each other retard. And that’s a good thing.
>>537069623>>537069901>>537069984China is 8 months behind USA not even close LOL
>>537069901This Code v3 benchmark is much worse because its narrowly focused on software engineering tasks only with limited datasets and potentially noisy/private evals. BridgeBench is broader, covering diverse reasoning, knowledge and capabilities for a more reliable overall ranking. GTFO from my thread, the retard Pole xD
>>537067941LMAO >cant run locally
>>537068975I know, pretty much a circlejerk. More optimized could also mean less flexible but see the point, in many cases that would just work equally with less demand.
>>537068710why aren't you using a Polish model? Oh wait....
>>537070177muzzies tell me Sam Altman and Dario is Polish
>>537070149>BridgeBenchdogshit jeet benchmarkcope harder
>>537070038KingBench 3 is even worse because its a narrow subjective YouTuber-made coding test (mostly one-shot frontend/app generation) with low rigor, no broad peer review and high variance. It doesnt test general reasoning/knowledge like BridgeBench, also failing to capture GLM-5.2s strengths in cost, speed and broad capabilities.
>>537070177I'm using SAR 3.1BASED Indian AI
>>537070332>saaaar my jeet benchmark good your american benchmark badthis is you
>>537070504Even if your shitty benchmark is right (not really), difference between GLM-5.2 and Fable 5 is minimal, not worth to pay Amerikikes x10-x15 higher price xD
>>537070625by minimal you means 10 -20% better for Americans than you are correcti will pay for Antrhopic because i am not poor
>>537070861Only a Polish imbecile would be dumb enough to pay x10 bigger price for pretty much the same product only to finance well-being of subhuman judeo-nigger-spic overconsuming Mutts. Thats your money tho, do what you want, you fkn idiot xD
>>537071034i make at least 20k more because i use AIyou are poor so go use dogshit china models
>>537071166>Calls someone else poor and brags about a measly $20kTop kek
>>537071411>morethis jeet can't read
>>537071411This idiot spams my every thread. Typical 90 IQ impoverished Pole larping as a rich man. Probably still lives with parents xD
>>537067941>>537067941thank you based china for making cheaper electronics, cheaper AI, and better AI now away from jewish control
>>537068840And? How come they are able to run their "stolen" models without all the bullshit American ones have? If they are running "stolen" models and removing the jewish aspects, I don't see the problem. It's called competition.
>>537071502you don't even own a computergo post more about gay anal sex you russian subhuman
>>537071476That's irrelevant. No rich person brags about $20k more. $20k ain't shit unless you live in Africa
Rare example of a wise Pole xD
>>537067941>because it was supposedly so smart that endangered the national security It wasn't just because it was so smart, it's because it is Recurrent-Depth Transformer model. This is a massive oversimplification, but basically this means that it loops it's reasoning chain of thought internally in latent space. Normally, thinking models generate their chains of thought in plain text in token space before they generate their final response, and this is what you can see when you use the "show thinking" option or whatever they call it.But with an RDT model, it iterates through this thinking internally. This makes it very powerful and effective, but it also means the thinking is all hidden inside the black box where humans can't see it. That's really what spooked everyone so much.
>>537071597>you don't even own a computerkeks and their vendor locked tech: https://www.youtube.com/watch?v=KrksBdWcZgQ
>>537067941The US government is right about goddamn everything and I am tired of pretending it isn't.Nobody believes these bullshit numbers though and Anthropic will still be 1000 more times more valuable than Deepsikh.
>>537072798Fable 5 wasnt banned for its speculated architecture, but over export controls triggered by a reported jailbreak (even minor vulnerabilities) forcing Anthropic to disable it globally, not for latent CoT. GLM-5.2 offers superior practical tech to Fable 5: massive MoE (744B total/40B active) + DeepSeek Sparse Attention for efficient 1M context, 300+ tok/s, strong agentic/coding performance at 1/10th cost, MIT-licensed, no regional restrictions. Thats all xD
>>537072492by we he means YOU a russian subhuman who earns 580USD a month KEKKKKKKK
Goyim are not allowed a model which gives sound financial advice. That model is reserved for the elite. It's as simple as that.
