DeepSeek V4 Flash on OpenCode Go has started feeling kind of dumb to me. It is occasional, which makes it harder to pin down. A response will run long and still seem to miss what I asked.
I first noticed it two or three days after DeepSeek changed the official pricing. The timing made me suspicious, but timing is all I have. I cannot connect that change to whatever OpenCode is serving.
The official V4 Flash 0731 feels noticeably different to me, and that is the part I cannot explain. I am treating this as an AI model routing question for now. ZenMux lets me send the same API request through different models and providers, so my next step is to compare the Go route, the official route, and a third route side by side instead of guessing from timing. Has anyone done a recent comparison and seen the same gap?
[link] [comments]