China open source AI just beat US models for the first time. It has never been more over to USA / The Jews and their closed source woke AI models.
>>539090899did we do this last year with deepseek?
>>539090899none of these charts show anything of value.why would you think speed matters above accuracy?
>>539090899nigga i need ram, it's a hardware problem at this point, not software
>>539091057these charts are benchmarks for accuracy
>>539091026no, Deepseek only destroys on price/quality. I argue that Deepseek is going to win as it's cheap af and more than good enough.
>>539091186>no!>okay, yeah. we did this propaganda push last year
>>539090899pressing 'x' to doubtthese open models arent being run properly 90% of the time (nobody has the proper resources even for this chinky shit), every dinky provider is charging as much as the premiers for what they actually give you. the post training is so bad they are all stuck as juniors at best. you want intermediate you always gonna need something trained properly not the chink shit. sorry not sorry
>>539091356what you are seeing is a cheap chink shit model thats been trained to perform well on those tests. THATS IT
>>539091026DeepSeek v4 Pro was actually proven to be retarded by a benchmark called DeepSWE. It was trained to pass the other tests in a rigged way.
>>539091409yeah probably. i use AI to code constantly. DeepSeek v4 Pro was unusable trash and slow too. i'm starting to wonder if Chinese actually are retarded compared to Whites and faked their country's IQ tests too.
>>539091548>DeepSWEhttps://github.com/datacurve-ai/deep-swe/issues/21People still fall for this shit?
>>539091548when the propaganda dropped last year (Andrew anglin even pushed it) I tried itit took all of 3 seconds to determine it was garbagenow they're trying to push kimi as an LLM that runs locally and outperforms high end modelsit's a complete fantasy
an LLM or image generator that can't criticize the chinese communist party but doesn't need to respect American copyright law and is otherwise lacking in nannystate safety guardrails is going to be infinitely more useful than the inverse.
>>539090899CheckedBut no. What good is a benchmark if we’re not AGI yet? It’s still the case that a human has to personally run and use this thing so that human’s experience with it will matter more than anything and that will expand beyond just programming ability into clarity, warmth, and formatting. Benchmarking is just a dick-measuring contest to attract investors with.
>>539091026Welcome to the age of endless fake chink propaganda. The bugmen make $2 an hour.
>>539091677GLM 5.2 works but it's really slow. The little cost per 1M tokens is nullified by the fact that it consumes many more million tokens for the same task than other frontier models. Sad. I will try Kimi 3 tho. I can't wait for liberating myself from ZOG's models.
>>539092200For a week or so I am enjoying Grok 4.5 at half the price on Cursor. Way better trade off than overpriced OpenAI or Anthropic models for coding. Kimi can wait.
>>539092200Claude shits on all of these other faggy models. Claude is intelligence.
>>539092383Yeah but Fable and even Opus 4.8 are too expensive. At the moment I use GPT 5.6 Sol for planning and Grok 4.5 for implementation.
>>539090899The people at deepseek have been quite silent lately, expect some happening soon.
>>539090899Americans would rather pay money for a car that goes 73 mph than have a free car that goes 70 mph.
>>539091246can't you read? Deepseek was always and still is way behind on accuracy compared to top of the line stuff. This Kim 3 model is probably way more expensive than Deepseek. Deepseek was always about being cheap and at the first it wasn't that good but when all models became... Like they are today, you start looking at the price more while the top models are doing whatever, trying to create AGI or something, which is just money embezzling bs.
>>539090899Chinese models are just as woke. They will lecture you about woke topics or refuse to answer if you're politically incorrect.
>>539092200>GLM 5.2 works but it's really slow. The little cost per 1M tokens is nullified by the fact that it consumes many more million tokens for the same task than other frontier models. Sad. I will try Kimi 3 tho. I can't wait for liberating myself from ZOG's models.Same scam as DeepSeek.Incredibly unusably slow.Uses way more tokens per request making the fact it "cost less per token" meaningless.
>>539090899Everyone's using Fable, Sol, or a small local model atm.>>539092497you should be anticipating something from google right now
