deepseek-v4.1POST
DeepSeek V4.1 Flash 1017 vs GPT-6 Astra and Opus 5: similar accuracy, far lower cost
On mini_swe_v2, V4.1 Flash 1017 scores about 77% at roughly ¥1.5 per task, versus GPT-6 Astra at 75%/¥60 and Opus 5 at 74%/¥80.
美股彭于晏｜PYY Research
@pyy_investArchived 12 days agoOriginally posted 2026.10.18
美股彭于晏｜PYY Research @pyy_investOriginally posted 2026.10.18
【DeepSeek V4.1 Flash-1017：同精度，成本差几十倍】 mini_swe_v2，Pass@1 vs 单任务成本（数据截至2026-09-22）： DeepSeek V4.1 Flash-1017：正确率约77%，单任务约¥1.5 GPT-6 Astra：约75%，约¥60 Claude Opus 5：约74%，约¥80 差不多的分，钱差一个数量级。开源这边还在往帕累托左上角推。
x.com/pyy_invest/status/2111702569290506682Archived 2026.10.18 16:34