>Opus 4.8 performance at $0.14/$0.28 priceHoly fuck, its actually happening. The second Deepseek moment wasnt Kimi K3, it was Deepseek again
Using chinese ai should be banned.
The benchmarks
>>109421344jew trump supporter spotted
>>109421313They're going to make a lot of fucking money from erp kek
>>109421313>not x—yai;dr
>>109421357Not Opus level but still pretty good
>>109421357Why does it rank below GLM 5.2 on Artificial Analysis? Smells benchmaxxed
>>109421392nigga im bot ai, im a real human bean
>>109422146>im bot ai
>>109421357That’s great, meanwhile in the real world it actually sucks.
>>109421357>BenchmaxxingI’m tired boss
more like deepsneed am i right fellas?
>>109422146And a real hero?
>>109421357Not saying DSV4 is bad butTerminalBench is such a bad fucking benchmarkIt always assumes you have perfect test cases, a perfect testable environment, etc.
>>109421344Thank God I'm not American
How do you jailbreak it
>>109421313>>109421357But their own benchmarks show them getting completely btfo by Anthropic's old Opus models. Huh? What did they accomplish?
>>109425508pricing, it's very important.
i put $10 into deepseek api back when v4 preview released and i was shovelling some code into v4 pro to find some bugs. still have $5 credits leftnow they release this flash which is better than old pro and like half the pricegod i love china
>>109421344nah, being jew on the other hand...
I use deepseek (expert, from the side) and gemini 3.1 pro for reviewing the code I write, and deepseek generally is a lot worse and both finding problems and understanding what it's looking at.
>>109421344>I fucking love being Jewish how did you know?as usual the Jew fears the red star
>>109425518Is their cheapness worth the massive downgrade in performance and results? Idk man, they're still so far behind Anthropic even by heir own metrics.
>>109421313Competing in AI with china is going to be like trying to compete with them in making steel screws, nuts and bolts.
>>109425656It's an all china competition if we're being honest nanodesu.
How good is it at erotic writing?
>>109425625Even by Anthropic's own official statements they are only about 4 months behind. Stop being melodramatic about massive downgrades when this shit didnt even exist not that long ago. Your little slop project isn't going to suffer because the model scored a few points less on an arbitrary benchmark when your prompts are all probably dogshit anyways seeing as how you sound like a retard anyways.
>>109426156On most benchmarks they are pretty significantly behind Opus 4.8 which is a months old model at this point. This is by their own metrics >>109421357Opus 4.8 is from March of this year (almost 5 months old) and it's still head and shoulders above Deepseek, once again by their own metrics. Stop riding Xi's microdick. Kimi is the only one that's truly on the heels of the big boys.
>>109421313literally how. I filled the token like 2 months ago and still have like 2 bucks in the tank
>>109421313I really hope this makes OpenAI fold.
okay coolhow do i run it?
>>109421344>amerigolems seethingGod I fucking love the chinese lmao
>>109426575have 2 rtx 5090
>>109421344You will never grow back your foreskin
>>109426575just download it? idiot
I struggle to find an use for paid deepseek, I've been using the chatbot onlyThough I struggle to see why a model that's "somewhat worse" can justify getting fucking robbed by anthropic. This shit works good enough for me without moaning.
>>109421344YOU MUST USE JEWISH AI INSTEAD, GOY!I’ll trust Chinese before I trust the others.
>>109426767God tier refactoring machine with zed for dotnet and c++ projects.
>>109425508DS4Flash only takes like 200GB vram to run, compared to an estimated 3,000GB for Opus
>>109421313Anyone feel extremely uneasy about this price point? If you think about it, essentially this means humanity's new baseline mental capabilities are at the level of DSV4-flash. Everyone in the world. I don't think we're ready for this yet.
>>109421313>Opus 4.8 performanceWhat does that even mean in that cargo cult tadbubble of yours?Have you realized yet that your war of the autocompletes doesn't fucking matter, you dumb fucking coolie?
>>109427653It means your dumb ass is entirely replaceable for pennies
>>109427554No. I'm too retarded to program and too retarded to understand the repercussions for any of this. All that matters is that it can create programs beyond my comprehension that abstract to a higher level that I can work with. Now everyone can be a low ranking production line worker slaving beneath the benevolent guidance of an engineer chatbot.
>>109427554There are people who are too dumb to even use an LLM, at least.
>>109427653>permanent underclass noisesdont worry you can still do manual labor I guess
>>109421344>we should stop using <this>!Why, because it works better? Welcome to the basics of tech competition, where a company one-ups a rival company for us retards to c0ns00m.
>>109425625The model is 55 times cheaper on input and 138 times cheaper on output. It's not performing 138 times as well.You didn't even have the chance to invest in these Enron tier scams yet and you're already coping as if you're all in.
>>109421344*rubbing intensifies*
>>109426549i set up a deepseek account to do a clients google reviews and put $10 bucks in. I think at the rate he gets reviews, it would only spend $0,84 cents per yearAI is too cheap now
>>109432284Based if true, OpenAI can't compete with that shit.
>>109421344The best part about Chinese AI is they usually release their weights so anyone with a sufficiently powerful machine can run it themselves.
American AI is already exposed. The question is now where to go next. Will American AI pivot to a space that China cannot or will not compete in like RPG optimization, even though they previously were trying to avoid this ? It may be the only way. The demand for companionship simulation and sexbots are frankly not going away, and enterprise use case is driven based on cost optimization, so if China cedes this area that is the last frontier of value that America can take over (and this also is one of the few fields where Grok is competitive).
>>109432864chinese models are good for rpg too. you can get them to do spicy shit with some tweaks. american corps are too cucked by christfags and karens. they will cuck everything with filters the minute some incel teenager offs himself again and their parents sue
>>109421357Opus 4.8 is still better across the board, but this is close.
>>109432864Chinese models are open source and the ban on RPGs applies only to users based in China. They're already the best for this use case. Those who want quality over speed and value will stick with GPT or Claude. There will always be a market of those who need the assurance of the best of the best, and they have the money to spend. The middle and low tier and nobody startups will stick with Kimi or Deepseek, sure, but government projects and serious multi billion dollar enterprises are going to go with the SOTA
>>109432864wtf are you talking about chinese models are the best for sex chats bc american ones are censored to hell >>109432882yeah, this is exactly why you will never get sex chats from openai or claude. amerigolem will sue because little jimmy kills himself and parents need a paydaychina bans it domestically but it doesnt matter for us userstheres no real justification for average user to spend money on more expensive apis. rpg? chinese have less censorship and are cheaper. coding? chinese models much cheaper and increasingly competitive performance, they can handle the bulk of it.
>>109432616Why even paid, local AI is the future. Only fools pay for the same power a gaming PC offers.
>>109432958With the prices of specs right now, local running is going to be a long way off for most people.
>>109432958you overestimate the ability of the average cattle to set up an local llm
>>109426549This is the price level AI should be at, not the ridiculous prices openAI, anthropic, etc. are chargingI am not paying $1+ per prompt, AI companies can go fuck themselves
>>109432958It's still too much hardware required for too little performanceIf I had $20k to throw away, yeah I'd be running a local LLM right now.When the required hardware for acceptable performance dropps 1-2 orders of magnitude, everyone will be running local AI instead of these big mainframe AI's.
>>109426549>>109432284It's subsidized by the CCP.
china is so ahead of us guise...........
>>109433255>if you reinterpret and redefine definitions and meanings, you can invert what reality isRight off the bat, you cannot "get basically the same results with more tokens" this is completely wrong. That is not how this works. The capability of an LLM is not a matter of how many tokens it's using. You cannot have a super genius LLM if you let it use infinity tokens.This incorrect definition is used by slimy word weasels like you to then invert reality and claim that OpenAI and Anthropic are aktshully 'more efficient'. Which they are not. They are more capable, but that capability is independent of token usage.A better definition of token efficiency is how much a single prompt costs you.If you prompt deepseek, it will cost you 1 centIf you prompt claude, it will cost you 1 dollarCurrent Deepseek will never give you an answer as good as claude does, no matter how many tokens you let it use, but 99% of your prompts just need 'good enough' responses. For example, if I want to know how narcan prevents an opioid overdose - I do not need some super intelligence to tell me. Deepseek's 1 cent answer is equivalent to Claude's 1 dollar answer.But claude costs 2 orders of magnitude more to get that answer than deepseek does.And that's real efficiency. How much I'm spending on this bulslhit every time I ask it a fucking question.So fuck off with your jewish bullshitting with redefining words and metrics. I know what the fuck actual efficiency is you slimy fucking rat.
>>109433360>fuck off with your jewish bullshitting with redefining words and metrics>let's define the efficiency of the model not by cost per task completed, but by how geniouser it is and chinas clearly geniouser
>>109433379shut the fuck up you sniveling rat weasel before I prompt claude to ram your favourite dildo ten miles up your aids ridden asshole
>>109433396see, even you choose the most efficient model
holy shit what level of ai psychosis even is this?https://files.catbox.moe/a0jdv2.txt>12,500 wordsthe schizos are gunna summon roko's basilisk.
>>109433379nta but your reading comprehension is subpar
>>109421313The only good thing about DeepSeek is its fucking over stupid people who trust the CCP lmao
Jewish century is cancelled. Good try though
>>109433360>For example, if I want to know how narcan prevents an opioid overdose - I do not need some super intelligence to tell me. Deepseek's 1 cent answer is equivalent to Claude's 1 dollar answer.This is a terrible example. That's the sort of question you could use a free AI for. I hope you're not actually paying for this kind of usage.
>>109425442It's not censored.
>>109433444what the fuck is this
>>109436376Sam Altman's wet dream.
I use it on openrouter and it says CoreWeave is hosting this. They literally take Jensen's money just so they can host chinkshit?
>>109436069>This is a terrible exampleand you're autistic.When you want the ai to read your code or anything like that you have to use an API model. Trying to nigger rig a free web model into something like copilot usually isn't worth the hassle and then you get rate limited or banned. If you want AI to read through a shitload of someone else's code and understand how it works so it can give you an overview, it doesn't require super intelligence but it would be nice to have it be cheap.
But anons on /g/ do it for free, thats $0.00 per million tokens.
>>109436474It costs your IQ, not wallet.
>>109432971Nah you can run V4F 0731 in memory with an ewaste tier Rome board and get ~17t/s. If you bought all the parts now at their current premium that's like 2000-3000 dollars.>t. Ebay ewaste connoisseur
>>109433046Youre retarded if you think you need 20k in hardware to run a 300B model. 2 sparks will run this at full precision pretty easily for 8k and there are even cheaper options that will also work well like I said in >>109437624
congratulations bro, anthropic have had mythos since februaryonly 6-8 months behind
>>109435143my reading comprehension is fine, your chinese cope is subpar
>>109433041And the funny part? One dollar per prompt is still not enough to put them in the green. Not even close
>>109421313Thank god im canadian
>>109438584>saaar
>>109421313tiananmen square 1989
>>109438591fucking kek
This is ccp(hussein!) propaganda. All Chinese ai is helping Iran win. This should be banned f to save us from the terrorists.
>>109421313This post is antisemetic
>>109438613Who gives a fuck about retarded students getting put down like the societal drains that they were? Their primary complaint was the CCP removing guaranteed sinecures in favor of actually pushing for the economic growth that has made China the powerhouse it is today. Face it—China is ascendant and the West is rotten.
>>109421357ooo wow so implessive, i try it out this morning it do the same ting the last deepseek does, sounds like an ai script bot. Very ching very chong
>>109439477it wasn't just students, those students killed a lot of PLA soldiers too. It made the PRC look incompetent.
