Chubby♨️@kimmonismus
68AI 编辑部评分,满分 100
2026-07-31 22:07· 23分钟前
跳到正文
AI 摘要

DeepSeek 计划在北京西北约 350 公里的乌兰察布新增 1GW 算力,在推进高效模型的同时大幅扩张计算能力,以继续角逐前沿。此前其已发布 DeepSeek-V4-Flash-0731 权重,采用 MIT 许可,含投机解码模块,以更小激活参数超越 V4-Pro(预览版)。

Oh and btw: DeepSeek is looking to add 1 GW worth of compute in Ulanqab, about 350 kilometers northwest of Beijing.

So they are not only relying on more efficient models, but of course also want to continue playing at the state-of-the-art frontier in the future, and are therefore massively expanding their compute capacity.

via Bloomberg

Chubby♨️DeepSeek v4 Flash 0731 weights were just published! Plus: - technical report - MIT license "DeepSeek-V4-Flash-0731 is the official release of DeepSeek-V4-Flash,...
Chubby♨️ · @kimmonismus · X·2026-07-31 22:07·23分钟前
在 X 看原推· x.com
AI 摘要

DeepSeek 计划在北京西北约 350 公里的乌兰察布新增 1GW 算力,在推进高效模型的同时大幅扩张计算能力,以继续角逐前沿。此前其已发布 DeepSeek-V4-Flash-0731 权重,采用 MIT 许可,含投机解码模块,以更小激活参数超越 V4-Pro(预览版)。

Oh and btw: DeepSeek is looking to add 1 GW worth of compute in Ulanqab, about 350 kilometers northwest of Beijing.

So they are not only relying on more efficient models, but of course also want to continue playing at the state-of-the-art frontier in the future, and are therefore massively expanding their compute capacity.

via Bloomberg

Chubby♨️DeepSeek v4 Flash 0731 weights were just published! Plus: - technical report - MIT license "DeepSeek-V4-Flash-0731 is the official release of DeepSeek-V4-Flash,...
在 X 查看原推x.com