Opencode vs Deepseek harness: my experience with Qwen 3.8 27b

A developer shared their experience comparing Opencode and Deepseek harnesses with Qwen 3.8 27b. They found Deepseek struggled with continuation prompts and easily got mixed up, even with button functionality, unlike Opencode. While Opencode's initial results were weaker and required more follow-up, its continuation prompts were described as "magic." The developer also noted that even top closed models make many errors in their complex tasks, questioning if Pi or Hermes would be better alternatives.