EP38 · DeepSeek V4 / Hermes Agent / 腾讯混元 Hy3 · 04.25 早报 episode artwork

EPISODE · Apr 24, 2026 · 19 MIN

EP38 · DeepSeek V4 / Hermes Agent / 腾讯混元 Hy3 · 04.25 早报

from BestBlogs

今日精讲DeepSeek-V4 预览版正式发布:百万上下文成标配,1.6 万亿参数普惠开源今日精讲 ①:时隔五个月 DeepSeek 一次甩出 V4-Pro(1.6T/49B)和 V4-Flash(284B/13B)两款 MoE 预览版,1M 上下文成为所有官方服务标配,并配上 CSA 4 倍 + HCA 128 倍的混合稀疏注意力,把长文 Agent 的算力和显存都压到 V3.2 的 27% 与 10%。Pro Agentic Coding 评测达到开源最佳,公司内部反馈优于 Sonnet 4.5、接近 Opus 4.6 非思考模式;Flash 在简单任务追平 Pro,且只需 13B 激活。定价分别是 Flash 0.14/0.28 美元、Pro 1.74/3.48 美元每百万 token,明确瞄准华为昇腾国产芯片。开源大模型的「能力 + 成本」前线被一次推到新位置。来自 Deep...去小宇宙查看完整单集简介前往小宇宙评论区与主播互动

NOW PLAYING

EP38 · DeepSeek V4 / Hermes Agent / 腾讯混元 Hy3 · 04.25 早报

0:00 19:12

No transcript for this episode yet

We transcribe on demand. Request one and we'll notify you when it's ready — usually under 10 minutes.

No similar episodes found.

No similar podcasts found.

Frequently Asked Questions

How long is this episode of BestBlogs?

This episode is 19 minutes long.

When was this BestBlogs episode published?

This episode was published on April 24, 2026.

What is this episode about?

今日精讲DeepSeek-V4 预览版正式发布:百万上下文成标配,1.6 万亿参数普惠开源今日精讲 ①:时隔五个月 DeepSeek 一次甩出 V4-Pro(1.6T/49B)和 V4-Flash(284B/13B)两款 MoE 预览版,1M 上下文成为所有官方服务标配,并配上 CSA 4 倍 + HCA 128 倍的混合稀疏注意力,把长文 Agent 的算力和显存都压到 V3.2 的 27% 与 10%。Pro Agentic Coding 评测达到开源最佳,公司内部反馈优于 Sonnet...

Can I download this BestBlogs episode?

Yes, you can download this episode by clicking the download button on the episode player, or subscribe to the podcast in your preferred podcast app for automatic downloads.
URL copied to clipboard!