同一游戏提示词下 Opus 5.5 四档 effort 对比
同一提示词、同一模型，仅改 effort 档位，从 4 分钟出原型到 1 小时自建 QA 团队，附并排视频。
What happens if you give Opus 5.5 the exact same game prompt and change only the effort level? Same model. Same task. Fresh session every time. Zero intervention. Low vs Medium vs High vs Xhigh. The difference was way bigger than I expected: LOW → 4m35s · $0.79 →Built a working game fast. MEDIUM → 14m31s · $2.38 → Started testing in Chrome, stress-testing FPS, fixing bugs and improving controls. HIGH → 31m43s · $6.04 → Turned it into a real product: better movement, bosses, richer combat, mobile polish, balancing, screenshots, performance checks. XHIGH → 1h17m · $9.35 → Basically became its own QA + game-design team. It created different player bots, realized its first bots were unrealistically good, redesigned the difficulty curve, retested, found UI/touch bugs, fixed them, and kept iterating until human-like bots died around the intended point. The simplest way I can describe it: >Low builds. >Medium tests. >High productizes. >Xhigh challenges its own work. So effort is not just “more thinking” It can completely change how the model behaves as an engineer. Video is the same prompt, side by side. 👇