Same Blender render prompt: DeepSeek platform 44m vs OpenCode Go 2h
With the same prompt, DSH harness and high reasoning effort, the official DeepSeek platform averaged 212 tok/s with 99.8% cache hits in 44m50s, while OpenCode Go took 2h05m at 170 tok/s.
🧵 ROUND 3: DeepSeek-v4.1-Flash-1017 - Same prompt, same harness (DSH), same reasoning effort (high), but two different providers. Time for a blender render this time! DeepSeek Platform (44m 50s): • Cache hit: 99.8% • Uncached input: 78,874 tok • Cached input: 32,812,416 tok • Output: 301,162 tok • Speed: 212 tok/s avg OpenCode Go (2h 5m 25s): • Cache hit: 99% • Uncached input: 276,498 tok • Cached input: 24,345,856 tok • Output: 185,155 tok • Speed: 170 tok/s avg What do you think did better?
This post quotes another post: “DeepSeek V4.1 Flash: DeepSeek Platform vs OpenCode Go on the same prompt”