The biggest problem with qwen3.8 is it emits way too many reasoning tokens, deepseek strikes a way better balance
Turn the reasoning/thinking/effort down?