Qwen isn't stopping !! (And trolling sama lol)

[-]

Own-Potential-2308@reddit

What even are these people's incentives to drop so much???

Reply

[-]

Prestigie nationally and internationally, and establish themselves as a leading AI lab, contrary to western perceptions. Remember when Altman tweeted something like its easy to copy but hard to do genuine innovation when DeepSeek V3 was released?

Reply

[-]

RobbinDeBank@reddit

Sam thinks OpenAI invented all of AI itself

Reply

[-]

RhubarbSimilar1683@reddit

He is kind of right through. ChatGPT is what started all of this

Reply

[-]

gkxk@reddit

"itself"

Reply

[-]

OrbitalOutlander@reddit

“quotation marks”

Reply

[-]

BoJackHorseMan53@reddit

Alibaba is one company of its kind (b2b seller), they have no competition. Businesses in every country in the world buy from Alibaba.

Reply

[-]

fufa_fafu@reddit

Alibaba is also trying to put AI into everything from car driving assist systems to robots to drones to their super app and more. Open source models generate tons of users that they can use for training data.

Reply

[-]

BoJackHorseMan53@reddit

Open source models run on users hardware which is not sent to Alibaba. They didn't have to open source their models to put them in cars or their super app. They would still get user data that way. No one is buying their cars or downloading their apps because of the open source AI models.

Reply

[-]

fufa_fafu@reddit

I think they're also putting those on the Qwen mobile app. Word of mouth and all.

Reply

[-]

BoJackHorseMan53@reddit

People of the world would use Qwen models even if it was closed source. A very tiny fraction of people have the hardware to run local models or care about local models.

Reply

[-]

MichaelXie4645@reddit

Don’t know, but hey, whoever don’t support this can have a friendly conversation with me.

Reply

[-]

agentspanda@reddit

Same reason China tries to excel at everything else- they want to monopolize the market then use their position to bend global business around their hegemony (and accordingly their values and their interests). Makes sense other companies aren’t playing since every open sourced model becomes fodder for China’s future endeavors. Personally I’m not a fan since I don’t live in China, but for them and their people I’m sure it’s going great (or as great as things go for their people).

Reply

[-]

fufa_fafu@reddit

>Same reason China tries to excel at everything else- they want to monopolize the market then use their position to bend global business around their hegemony (and accordingly their values and their interests). Are you describing china or the united slaves of america here? Thanks to BYD my relatives in southeast asia can get cheap, affordable, and reliable EVs. Thanks to Qwen I run local models that's been a lifesaver for my work. Thanks to every chinese company selling on marketplaces I get hidden gems for bargain prices. american capitalism is unsustainable and is well on its way to die.

Reply

[-]

BFGsuno@reddit

That's how progress is made. Some people don't need idiotic competition.

Reply

[-]

Hugogs10@reddit

Except competition is clearly driving a lot of this lol

Reply

[-]

BFGsuno@reddit

american companies who spends $trilion+ and they need something to show investors ? sure. But they are the one losing to chinese R&D companies operating on fraction of budget, on fraction of hardware which release most of the models for free with mostly apache license...

Reply

[-]

rkoy1234@reddit

> fraction of the budget idk where this myth is coming from. their ai space is humongous, and they have far more institutions (both corporate and academic) slaving away publishing research papers on AI on an hourly basis. US's publication numbers are laughable compared to china these days. sure, the individual salaries of engineers/researchers might be lower, but AI is a national project backed by both financial and logistical support in china, with extremely competitive subsidies, benefits, and exemptions. Idk why we're all so eager to categorize them as some backyard engineers. China is going all in - IMO much more than the US at this point. US really needs to pour *more* resources at a national level if it wants to remain competitive.

Reply

[-]

BFGsuno@reddit

dude, R1 was trained like on fraction of what xAI or OpenAI, Meta have.

Reply

[-]

DorphinPack@reddit

Competition is also what created the closed source silos that occasionally fart out weights. Who knows what advancements we’d have if the most cutthroat didn’t just wipe the field

Reply

[-]

BoJackHorseMan53@reddit

Competition doesn't seem to push America to release open source models.

Reply

[-]

Illustrious-Lake2603@reddit

Because they want GPT5 dead

Reply

[-]

Arcosim@reddit

My wet dream is that Altman drops GTP-5 and the next day DeepSeek drops R2 and kills all the hype.

Reply

[-]

cpsnow@reddit

Dropping it 3 days before would be even more so amazing.

Reply

[-]

Familiar_Gas_1487@reddit

You got a Time Machine?

Reply

[-]

EndStorm@reddit

This is what I want to happen lol.

Reply

[-]

ThinkExtension2328@reddit

All hail king Qwen 😂😂😂, man my popcorn needs a refill.

Reply

[-]

eposnix@reddit

Notice they haven't come out with a model superior to the Sota... ever. They need GPT-5 to distill from, just like all the other Chinese labs.

Reply

[-]

BoJackHorseMan53@reddit

Contrary to popular belief humans aren't driven by carrot (money) and stick (starvation) that's for donkeys.

Reply

[-]

Important_Concept967@reddit

Of course they are, They just aren't the only things

Reply

[-]

BoJackHorseMan53@reddit

Didn't say some people aren't any better than donkeys

Reply

[-]

woahdudee2a@reddit

they feed on sam altman's tears

Reply

[-]

Creative-Size2658@reddit

Aren't we all?

Reply

[-]

Tricky-Appointment-5@reddit

Dont we all?

Reply

[-]

paryska99@reddit

Donut we all?

Reply

[-]

MealFew8619@reddit

So say we all

Reply

[-]

EndStorm@reddit

Spin up the FTL drive, we've got some tears to witness!

Reply

[-]

mnt_brain@reddit

CCP is dumping an absolute fuckton amount of pressure and funding for these teams so they’re producing at light speed

Reply

[-]

mister2d@reddit

One reason is because it reduces the dependency on closed source APIs. Another could be to enhance brand reputation.

Reply

[-]

ThenExtension9196@reddit

The incentive is to get recognition and then get a 100m job offer from Meta and move to the US and live like luxury.

Reply

[-]

InsideYork@reddit

The incentive system in China, versus the US’s. I liked the incentives of France too.

Reply

[-]

martinerous@reddit

I like "small-but-good" better than "state-of-the-art". The second one makes my eyes roll painfully. Almost every article has "SOTA" and it's become meaningless.

Reply

[-]

entsnack@reddit

81.7% on AIME25 lmao, so much for trolling

Reply

[-]

Creative-Size2658@reddit

https://artificialanalysis.ai/evaluations/aime-2025 Qwen3 got 91.0%, better than O4-mini (90.7%) So, it looks like a good trolling to me...

Reply

[-]

LuciusCentauri@reddit

That 91% is the 235B model. 81.7% for a 4B model that can run on your phone is pretty decent tho

Reply

[-]

InsideYork@reddit

Wonder if this will this be better than gemma3 2n

Reply

[-]

Creative-Size2658@reddit

I didn't see any reference to gemma3 2n (here)[https://artificialanalysis.ai/evaluations/aime-2025?models=gemma-3n-e4b], only gemma 3n e4b, and it's not good. Only 14.3%

Reply

[-]

InsideYork@reddit

I mean for general use. I don’t think 2n is good, but for running on phones the priorities will be different.

Reply

[-]

Creative-Size2658@reddit

I wasn't sure. I don't know, but that's an interesting question. What capability would you say is more important on a phone? I imagine something able to call some functions, and easily "learn" how to call some others. And maybe answering some basic questions reliably enough. IMO, stuff like creative writing and conversation skills wouldn't be very useful on a phone. Probably more in video games though.

Reply

[-]

InsideYork@reddit

I think knowing how to do things with apps, reminding you and probably creative writing too. A helper mode too.

Reply

[-]

Creative-Size2658@reddit

Yeah that makes sense.

Reply

[-]

Creative-Size2658@reddit

WTF, I didn't even know this 4B model was doing 81.3%! I just saw the benchmarks on their HF page. So it's even better than GPT-OSS 20B (78.7%) Nice.

Reply

[-]

NeuralNakama@reddit

Jokes aside, I hope they announce the VL models.

Reply

[-]

Rich_Artist_8327@reddit

I think americanos should give up, we europeans even didnt start :)

Reply

[-]

Competitive_Ideal866@reddit

> EDIT: forgot Mistral. And Google Deepmind (Gemini, Gemma, Genie, AlphaGo, AlphaGeometry, ...).

Reply

[-]

Rich_Artist_8327@reddit

are they europeans?

Reply

[-]

Competitive_Ideal866@reddit

I'm sure the team are from all over the world but they are based in London.

Reply

[-]

4orth@reddit

Deepmind was London based originally if I'm rendering correctly.

Reply

[-]

FrontLanguage6036@reddit

Imagine if they drop a reasoning model as good as gpt oss 20b. But realistically, it would be either qwen-image small or something for qwen coder (maybe?)

Reply

[-]