DeepSeek V4.1 Flash 1017 同题双供应商实测：官方平台 213 tok/s vs OpenCode
同一提示词、同一 DSH harness、同等推理强度下对比两家供应商：官方平台 1h30m、缓存命中 99.8%、213 tok/s；OpenCode Go 1h18m、98%、143 tok/s。
🧵 FINAL ROUND: DeepSeek-v4.1-Flash-1017 - Same prompt, same harness (DSH), same reasoning effort (high), but two different providers. I'll be making another final post comparing the stats, my thoughts in details, and whether the OpenCode Go subscription is worth it still. DeepSeek Platform (1h 30m 42s): • Cache hit: 99.8% • Uncached input: 91,329 tok • Cached input: 43,253,376 tok • Output: 265,159 tok • Speed: 213 tok/s avg OpenCode Go (1h 18m 31s): • Cache hit: 98% • Uncached input: 182,546 tok • Cached input: 8,274,688 tok • Output: 77,621 tok • Speed: 143 tok/s avg What do you think did better?
这条是引用着另一条写的：「DeepSeek V4.1 Flash 同题双平台耗时与缓存对比」