This literally made my day! With the right orchestration, it can do >95% of what Opus/GPT5.5 can do at 2.5% the cost, extremely impressive model. This is the best way you can feel 'tokens too cheap to meter' I'm creeping up to my first 1b token day!

@peterom I’ve been running a quantized ds4 flash on the spark and it’s been pretty fantastic for a local model!

@SyntaxDiffusion Wow, that's amazing! I was looking up local hardware for it in case of an apocalypse and was curious was ds4 compared