Hard to argue against DSV4-Flash being the best cost to ratio LLM. It's almost @ GLM-5.2 on engineering & general intelligence, while being significantly cheaper than even 5.6-Luna.

Demand continues to be off the chart My cline sessions are noticeably slower than 1st few days. I continue to run it overnight while I go to sleep. The long context sessions are great for SWE work while being general productivity gains. If DS can continue to scale context to like 10m token or infinite, then utilit — via @tphuang

Cost-efficient DSV4-Flash