OpenCode 宣布为 DeepSeek V4.1 Flash 永久提供 60 美元额度

OpenCode 启动"Operation Cheepseek"第二阶段，宣布将 DeepSeek V4.1 Flash 在 OpenCode Go 中的 60 美元使用额度从限时调整为永久有效1。

DeepSeek V4.1 Flash 采用 5520 亿参数的 MoE 架构1，其中输入侧激活参数量为 80 亿，输出侧为 160 亿1。相比上一代模型，该版本显著降低了内存需求，KV Cache 对 HBM 的需求降低至四分之一，对 SSD 的需求降低至八分之一1。截至 9 月 25 日，DeepSeek V4.1 Flash 在 OpenCode 平台近期使用量中排名第一，占约 13% 的 200 万级使用量1。

OpenCode has announced the second phase of "Operation Cheepseek," converting the previously time-limited $60 credit allocation for DeepSeek V4.1 Flash on OpenCode Go into a permanent offering 1. The $60 credit is equivalent to approximately 403.5 Chinese yuan 1.

DeepSeek V4.1 Flash utilizes a mixture-of-experts architecture with 552 billion parameters, featuring 8 billion activated parameters on the input side and 16 billion on the output side 1. The model has achieved significant efficiency improvements compared to its predecessor, reducing KV Cache demands on high-bandwidth memory to one-quarter of previous levels and on solid-state storage to one-eighth 1. As of September 25th, DeepSeek V4.1 Flash ranked first in recent usage on the OpenCode platform, accounting for approximately 13% of the platform's 2-million-level usage volume 1.