Leeroy

Anyone knows why @OpenDesignHQ's DeepSeek thinks and outputs more even though (based on my benchmarks) it doesn't seem to be quantized compared to OpenCode? The screenshot below is an average of 9(x2) benchmark runs, each taking more than an hour.

Dear open model labs, if you’re gonna release the weights anyway, why spend so much time censoring the model when some random guy is just gonna uncensor it in a weekend can we skip this part please