[UncleOxidant]

Run Qwen3-coder-next locally. That's what I'm doing (using LMstudio). It's actually a surprisingly capable model. I've had it working on some LLVM-IR manipulation and microcode generation for a kind of VLIW custom processor. I've been pleasantly surprised that it can handle this (LLVM is not easy) - there are also verilog code that define the processor's behavior that it reads to determine the microcode format and expected processor behavior. When I do hit something that it seems to struggle with I can go over to antigravity and get some free Gemini 3 flash usage.

[zirror]

What kind of hardware do you run it on?

[UncleOxidant]

Framework Desktop (AMD Strix Halo with 128GB). Runs it around 27 tok/sec which is quite acceptable.

[kristianpaul]

Same here