@Alibaba_Qwen@LebrelAI Qwen 4 max is closed so it must compete with closed models. Q4max gets beat by gpt 6.1 terra. You dont get to garner community goodwill by parading youselves as open source champions but keeping your frontier models closed.
Its a terrible release, slow, overthinks AND pricier!
@melvinjohnsonp more usage and i want the speed to be same unlike what you did with 3.8 flash. 3.8f was fast at launch but impossibly slow 2months later.
@jauneclaude because you dont understand what MoE is. You dont have "python expersts" or "spanish experts". A typical query might activate 80/100 experts of the model so you are still swapping multiple GBs for each token. With experts on SSD you'd be looking at 1-2t/s.
@sundarpichai@GoogleAIStudio Just give us gemini 4 flash. The current 3.8 flash is slower, more expensive AND dumber than deepseek v4.1 flash 1017. Argon is just a novelty for Pro users because of its heavily constricted usage limits. I expect carbon to be the same.
@alpaimdev if you cant handle hate, dont post on the internet. Dissenters will always have a higher probability of commenting than those who like the thing.
@BenGreenJeru@washingtonpost@JonathanLanday that doesnt disprove the Gaza genocide wasnt a genocide....
but population increased in the 1994 genocide against the Tutsi in Rwanda
@BenGreenJeru@washingtonpost@JonathanLanday "Genocide is the intentional and systematic destruction, in whole or in part, of a national, ethnic, racial, or religious group."
population can increase if number of people killed are less than being born