Armen Hovhannisyan

Hi I am Armenian and I am against agression

[Armen Hovhannisyan @Armen181]

@PeasantSmith I guess some good cookies are coming for v100s 😅

[Armen Hovhannisyan @Armen181]

@PrideOfBlair @filicroval Then it is a matter of hardware to make it fast, or the qwen3.6-35B-A3B model is fast, so is it flash?

[Armen Hovhannisyan @Armen181]

@QwenDevs @KyleHessling1 I hope we can same for qwen4-flash

[Armen Hovhannisyan @Armen181]

@MiniMax_AI Is there any information about model size?

[Armen Hovhannisyan @Armen181]

@PeasantSmith

[Armen Hovhannisyan @Armen181]

@CommandCodeAI when GLM5.3 and deepseek v4.1?

[Armen Hovhannisyan @Armen181]

@Da7_Tech When qwen3.8 27b beats Opus 4.8? I guess you meant Opus4.6.

[Armen Hovhannisyan @Armen181]

@Da7_Tech Just need to have computer, 20K$

[Armen Hovhannisyan @Armen181]

@CommandCodeAI Any update from Deepseek or glm5.3 being free?

[Armen Hovhannisyan @Armen181]

@PeasantSmith If magic is already happening, then I wish 1400t/s for prefil.

[Armen Hovhannisyan @Armen181]

@ItsmeAjayKV It is always about user data, even if you have opus5.5 generated code that will help DeepSeek and GLM become smarter and cheaper

[Armen Hovhannisyan @Armen181]

@_ryu15_ Yes we can but now tokens cheaper then electricity. It depends what you are developing

[Armen Hovhannisyan @Armen181]

@_ryu15_ Nvidia: I need your data

[Armen Hovhannisyan @Armen181]

@opencode hard to say is it good idea or not, now none of us will have acceptable service.

[Armen Hovhannisyan @Armen181]

@bindureddy In practice you want to spend a little more and use glm5.3-flash, it is worth.

[Armen Hovhannisyan @Armen181]

@PeasantSmith @axitek @needmorevram V100 users

[Armen Hovhannisyan @Armen181]

@tjm8874 I don't think that the average desktop user needs local AI, use some 10$ subscription or multiple of them, when you understand why local AI is needed, then you are already not an average user.

[Armen Hovhannisyan @Armen181]

@chasmmmmmmmmmmm I'm expecting Qwen3.8 Flash Coder to run at around 40-50 t/s with the Strata engine, using CPU, GPU, RAM, and SSD for better performance.