你写代码调模型时，试试在model前加个ship-like/，据说能省一半钱，功能和效果丝毫不差。适合追求性价比的开发者。
通过将代码中model参数替换为ship-like/前缀（如model='ship-like/your-favorite-model'），即可使用智能路由服务。该服务承诺相比直接连接模型节省50%成本，同时保持100%的能力和行为一致性。其核心是在收到请求后动态决定计算量、模型组合、工具和执行方式。代价是增加额外延迟，但对多数开发者来说性价比很高。
Huge if this works as promised! Replace "model='y…
Huge if this works as promised!
Replace "model='your-favorite-model'" with "model='ship-like/your-favorite-model'" in your code.
You don't need to change anything else and still get three guarantees:
1. Pay 50% less money than connecting directly to the model 2. Enjoy 100% the same capabilities 3. Enjoy 100% the same behavior
Apparently, they do this using some sort of "smart routing" behind the scenes.
After getting a request, they can determine:
1. How much compute to spend 2. Which model (or combination of models) to use 3. What tools, harnesses, ensembles, etc. to use 4. Which execution method to use
I'm going to guess the tradeoff here would be extra latency when calling the model, but at 50% the cost, it looks like a no-brainer for most people.