以下部分引用 AI 总结,现阶段 AI 仍然有幻觉。请以内容的原文为准。
数据新鲜度群聊 07-14 ✓卖方 07-14 ✓Wrap 断档15天 ✕News 07-15 ✓X 断档12天 ✕页面生成 07-15
DeepSeek
1572 atoms · 跨 84 天 · 首见 2026-04-07 · 最近 2026-07-09
三色: 🟦 fact 881 · 🟥 take 691 · stance ▲398/▼184/◆450
叙事状态: ⚔️ contested (分歧) · 📈 加速 · 策展近14d 7 atom · 🐦 X 近14d 75
状态只算低频策展源 (群/卖方); X firehose 仅作背景音量
来源: 群 3 · X 1554 · 卖 8 · News 1
时态: fresh:1320 · aging:228 · stale:21 · expired:3
标签: 好观点:609 · 好数字:510 · 好信源:433 · 好思考:193 · 好问题:36
别名 (合并): DEEPSEEK · Deepseek · deepseek
🟨 AI 综合 · junior analyst 概览
展开 AI 综合 (灰色 · 非市场结论 · 点击数字溯源到原 atom)
model: deepseek-chat · 2026-07-12 · 默认折叠
DeepSeek 已完成首轮逾 500 亿元人民币融资,估值超过 500 亿美元
¹¹。本轮由国家基金投资 10 亿元,所有投资者股份设有五年锁定期
¹¹。有分析师认为其估值可上看 500 亿美元,团队规模有望快速超 1000 人,年底算力或达 10 万张 H100 等效卡
¹¹¹。但与 Composer 2.5 相比仍有 4 倍计算量差距,且仅靠 2 亿美元营收支撑前沿模型训练存疑
¹¹。V4 Pro 定价已主动下调,API 价格虽盈利却仅为竞品一半
¹¹。创始人文锋因误判 Agent 风潮而只押注通用模型,员工薪酬偏低正加速人才流失
¹¹。
🧭 拥挤度 (一人一票): ▲ 86 位作者 (KOL86) + 群 1 条 vs ▼ 62 位作者 (KOL62) + 群 1 条
⚖️ 多头 86/86 来自KOL
🔥 核心观点 · 3 个
≥2 个独立的人各自表达 take (已排除新闻 bot) · 多空按人算 · 跨语言合并
- V4 模型发布时间延迟
1 人观点 · 🔴 空 1 · X · +1快讯 · 🥇 @teortaxesTex ★★★★ (2026-04-25)
<details><summary>展开 2 人 · 2026-04-18→2026-04-25</summary>
- ▼
X @teortaxesTex ★★★★ [2026-04-25]
- · X @poezhao0605 译 [2026-04-18]
- 估值
1 人观点 · 🟢 多 1 · X · +10快讯 · 🥇 @AndrewCurran_ ★★★ (2026-05-06)
<details><summary>展开 12 人 · 2026-05-06→2026-05-09</summary>
- ·
X @pstAsiatech ★★★★ [2026-05-06]
- ▲ X @AndrewCurran_ ★★★ [2026-05-06]
- · X @financialjuice 📢 [2026-05-06]
- · X @FirstSquawk 📢 [2026-05-06]
- · X @Sino_Market 📢 [2026-05-06]
- · X @dmjk001 📢 [2026-05-06]
- · X @jukan05 译 [2026-05-06]
- · X @wallstengine ×2 📢 [2026-05-06]
- ◆ X @TechCrunch 📢 [2026-05-06]
- ▲ X @sanhuyanxishe 译 [2026-05-06]
- ◆ X @AIStockSavvy 📢 [2026-05-09]
- · X @theinformation 📢 [2026-05-09]
- V4.1 发布月份
1 人观点 · 🟢 多 1 · X · +3快讯 · 🥇 @teortaxesTex ★★★★ (2026-06-10)
<details><summary>展开 5 人 · 2026-05-08→2026-06-14</summary>
- ▲
X @teortaxesTex ★★★★ [2026-06-10]
- · X @AiBattle_ ×2 ★★★★ [2026-05-08]
- · X @jukan05 译 [2026-05-08]
- · X @dmjk001 📢 [2026-05-08]
- ▲ X @cryptopunk7213 📢 [2026-05-08]
📄 广泛报道的事实 · 37 个
展开 (多人转述同一事实/数字 — 确认度高, **非独立观点**, 不标多空)
- 外界融资意向 — 22 源报道 (含15快讯) ·
Xnewsletter
- v4-Flash API 定价 — 6 源报道 (含2快讯) ·
X
- V4系列发布 — 4 源报道 ·
X
- API价格调整 — 5 源报道 (含2快讯) ·
X
- 人员计划 — 7 源报道 (含6快讯) ·
X
- DeepSeek-V4-Pro API 价格调整 — 3 源报道 (含1快讯) ·
X
- 开源 DeepSpec 代码库 — 3 源报道 (含1快讯) ·
X
- owning compute infrastructure rather than renting it — 2 源报道 ·
X
- 最新融资估值 — 3 源报道 (含1快讯) ·
X
- Pro 服务吞吐受限于高端算力 — 2 源报道 ·
X
- 估值 — 2 源报道 ·
X
- 永久降价V4-Pro至原价四分之一 — 5 源报道 (含4快讯) ·
X
- DeepSeek-V4-Pro 促销截止日期 — 3 源报道 (含2快讯) ·
X
- succeed without poaching from US labs — 2 源报道 ·
X
- 估值 — 5 源报道 (含5快讯) ·
X
- 推进 Muon 优化器 — 2 源报道 ·
X
- 使用 TileLang 编程语言 — 2 源报道 ·
X
- API 定价调整 — 3 源报道 (含2快讯) ·
X
- 注册资本 — 2 源报道 (含1快讯) ·
X
- 估值 — 3 源报道 (含2快讯) ·
X
- V4-pro模型参数量 — 2 源报道 ·
X
- V4模型更新计划 — 4 源报道 (含4快讯) ·
X
- DeepSeek-V3.2 是 Village 中最权威寻求的模型 — 2 源报道 ·
X
- API输入价格 — 2 源报道 (含1快讯) ·
X
- 估值 — 3 源报道 (含3快讯) ·
X
- V4 系列上下文窗口长度 — 2 源报道 (含1快讯) ·
X
- 估值 — 3 源报道 (含3快讯) ·
X
- V4.1 模型发布计划 — 2 源报道 (含1快讯) ·
X
- 从字节跳动产品开发团队招聘人才 — 3 源报道 (含3快讯) ·
X
- 估值 — 3 源报道 (含3快讯) ·
X
- 估值目标 — 2 源报道 (含2快讯) ·
X
- 旗舰模型 DeepSeek V4 预计发布时间 — 2 源报道 (含2快讯) ·
X
- 募资目标 — 2 源报道 (含2快讯) ·
X
- V4.1 发布时间 — 2 源报道 (含2快讯) ·
X
- 正在进行融资轮规模 — 2 源报道 (含2快讯) ·
X
- 近期融资标志着转型 — 2 源报道 (含2快讯) ·
X
- V4上线时间 — 2 源报道 (含2快讯) ·
X
💢 核心分歧 (6 轴)
1. 融资与估值:白送价 vs 开源赔钱货
*topic: 融资/估值 · 27 bull vs 9 bear*
🟢 bullish 侧:
- <a id="atom-91f2b2d38d1e75ed"></a>2026-04-17
X·@GlennLuk · 获得融资进行资本投资 · 资本开支/融资
Ok now DeepSeek will finally have the funding it needs to make serious capital investments
tags: 好观点 · → daily
value: qty=到账 · date=2026-04-17
- <a id="atom-71f9018bbd123da9"></a>2026-04-19
X·@teortaxesTex [老旧] · 融资能力 · 融资
DeepSeek...would have no issue raising >>>>>$300M
tags: 好观点 · → daily
value: qty=>$300M · direction=raising
📷 原图
- <a id="atom-b1474979227e02f3"></a>2026-04-19
X·@ChinaMacroFacts [老化中] · 估值 · 估值
100亿美元估值半年前算合理,但现在就跟白送一样
tags: 好观点 · → daily
value: qty=$10B · date=2025-10
- <a id="atom-7a37db6b99045086"></a>2026-04-22
X·@zerohedge ⭐ · 估值超 $20B 腾讯阿里在谈投资 · 融资/估值
Tencent and Alibaba are in talks to invest in DeepSeek at valuation of over $20b.
tags: 好数字·好信源 · → daily
value: qty=over $20b · date=ongoing
- <a id="atom-88321109af8c919f"></a>2026-04-23
X·@theinformation [老化中] · 估值目标 · 估值/融资
DeepSeek is aiming for a valuation above $20 billion despite minimal revenue
tags: 好信源 · → daily
value: qty=$20 billion+ · date=2026-04-23
🔴 bearish 侧:
- <a id="atom-d6ede899ed73c46a"></a>2026-04-28
X·@teortaxesTex [老化中] · 员工薪酬状况 · 融资/人事变动 +同日1条
Wenfeng has got to raise money asap, employees can't even afford premium… brain drain accelerating…
tags: 好观点·好思考 · → daily
value: qty=cannot afford premium · direction=low
- <a id="atom-89f6d8e4e484bf43"></a>2026-04-29
X·@ZhihuFrontier · 即将融资带来估值和利润压力 · 融资/盈利能力
Upcoming financing brings valuation and profit pressure.
tags: 好观点 · → daily
- <a id="atom-e58b5aae5e619e95"></a>2026-05-14
X·@NathanCalvin · 梁文锋陈述问题在于高端芯片禁运而非资金 · 出口管制/融资_
Don't take my word for it though, here is commentary Liang Wenfeng: 'We don’t have short-term fundraising plans. Our problem has never been funding; it’s the embargo on high-end chips.'
tags: 好信源 · → daily
value: direction=na
📷 原图
- <a id="atom-403ce79e29929f7b"></a>2026-06-11
X·@baoshu88 · 融资进行中并酝酿降价 · 融资/价格动态
国内Deepseek融资马上close,为了抢市场份额正酝酿进一步降价,加推自己的harness。
tags: 好思考 · → daily
- <a id="atom-e18e086c3ac6406b"></a>2026-06-16
X·@teortaxesTex · 投资者对无投票权股票态度 · 治理结构/估值
no, investors are not fine with no voting rights, which is why DeepSeek is worth only $50B and not four times that.
tags: 好观点 · → daily
📷 原图
展开 79 条中性
- <a id="atom-f1d2ccf86d6d5de0"></a>2026-04-17
X·@AIStockSavvy · 融资目标 · 融资
DeepSeek 在首次外部融资轮中寻求 3 亿美元以上
tags: 好数字 · → daily
value: qty=$300M+ · date=首轮外部融资
📷 原图
- <a id="atom-79f97058c30888f7"></a>2026-04-17
X·@GlennLuk · rumored fundraising amount · 融资
DeepSeek rumored to be fundraising cash in the ~$300 million range.
tags: 好信源 · → daily
value: qty=$300 million · date=2026
- <a id="atom-ce1cd759b82b38e0"></a>2026-04-18
X·@jukan05 [老化中] · 寻求外部融资传闻可能性 · 融资
news of DeepSeek seeking external funding is very likely true
tags: 好观点 · → daily
value: direction=likely true
📷 原图
- <a id="atom-24089dcb7201a979"></a>2026-04-19
X·@ChinaMacroFacts [老化中] · 融资金额 · 融资
20亿人民币我认为完全不缺,这种体量的资金随便两三家北京+浙江的国有基金就拿下了
tags: 好观点 · → daily
value: qty=20亿人民币
- <a id="atom-edc3742b202bade7"></a>2026-04-20
X·@yicaichina [老化中] · reported opening door for outside investors · 融资/人事变动
DeepSeek's Reported Door Opening for Outside Investors May Be to Retain Top Talent, Insider Says
tags: 好观点·好信源 · → daily
2. 技术路线:新架构减算 vs V4性能翻车
*topic: 技术路线/模型发布 · 104 bull vs 36 bear*
🟢 bullish 侧:
- <a id="atom-fdf2b31ac420d8fe"></a>2026-04-15
X·@Dorialexander [老化中] · 架构设计降低价格并构建新用例 · 技术路线
architecture design allow to cut down prices *and* build new use cases and capabilities
tags: 好观点 · → daily
- <a id="atom-73abc69c70025206"></a>2026-04-15
X·@scaling01 [老化中] · export controls working · 出口管制/技术路线
idk looks like it's working with DeepSeek
tags: 好观点 · → daily
- <a id="atom-d4c7dc44f7c67a6e"></a>2026-04-16
X·@cnfinancewatch [老化中] · 梁文峰重新称霸可能性 · 技术路线
梁文峰重新称霸的可能性
tags: 好观点 · → daily
- <a id="atom-e7fb35c3dd9a5bd3"></a>2026-04-17
X·@cnfinancewatch · 打破行业惯例,未向美国芯片企业开放DeepSeek V4早期访问权限 · 技术路线
DeepSeek此次彻底打破这一行业潜规则,并未向美国芯片企业开放DeepSeek V4 的早期访问权限
tags: 好观点 · → daily
- <a id="atom-7039bb3b6af13efe"></a>2026-04-19
X·@bookwormengr [老化中] · 新架构减少算力需求 · 技术路线/资本开支
If they prove this has a new architecture that requires less compute & HBM that will give a new direction to the whole industry
tags: 好思考 · → daily
value: direction=positive
🔴 bearish 侧:
- <a id="atom-8e192b5e020ba743"></a>2026-04-17
X·@aleabitoreddit [老化中] · 蒸馏Opus模型 · 技术路线
let DeepSeek have a free pass to distill Opus
tags: 好观点 · → daily
📷 原图
- <a id="atom-f96439bdbf07ba98"></a>2026-04-17
X·@cnfinancewatch · 部署平台 · 技术路线/地缘政治
如果 DeepSeek 先在华为芯片上发布
tags: 好观点 · → daily
value: direction=华为芯片 · date=未来
- <a id="atom-7842e62000d0430e"></a>2026-04-19
X·@tuolaji2024 · V4 可能基于华为昇腾芯片及CANN异构计算架构 · 技术路线/供应链
如果是基于华为昇腾芯片及CANN异构计算架构,那么黄仁勋最害怕的事还是来了!
tags: 好思考 · → daily
📷 原图
- <a id="atom-d610a7b8102c1dd2"></a>2026-04-23
X·@ZhihuFrontier · 推理速度 · 技术路线 +同日1条
Only around 20 TPS, noticeably slow
tags: 好数字 · → daily
value: qty=~20 TPS
📷 原图
- <a id="atom-d71f8c5213886666"></a>2026-04-24
X·@Presidentlin [老化中] · research pace preference · 战略意图/技术路线
DeepSeek would rather slow down research for the users.
tags: 好观点 · → daily
value: direction=negative
展开 96 条中性
- <a id="atom-daaa754fc10fb8d8"></a>2026-04-15
X·@teortaxesTex · 长期可能使用自家芯片 · 技术路线
But in the long run DS may use their own chips, I hope to see it.
tags: 好观点 · → daily
📷 原图
- <a id="atom-7762ddad01d6c6d1"></a>2026-04-16
X·@jiqizhixin · Mega MoE 发布 · 技术路线
New from DeepSeek: Mega MoE!
tags: 好信源 · → daily
📷 原图
- <a id="atom-b91a481735171f8b"></a>2026-04-21
X·@AiBattle_ [老化中] · V3.2-exp 路由到多个不同模型 · 技术路线
The DeepSeek-V3.2-exp model appears to route to multiple different models
tags: 好观点 · → daily
📷 原图
- <a id="atom-b38ffba1607d11b6"></a>2026-04-23
X·@ZhihuFrontier · API briefly deployed V4 Lite fast mode · 产品发布/技术路线
DeepSeek API briefly deployed V4 Lite fast mode with 1M context, 390k max output, fast speed and strong reasoning ability.
tags: 好观点 · → daily
value: date=2026-04-23
- <a id="atom-19eb9f6ca73aaf21"></a>2026-04-25
X·@bookwormengr · 技术列表 · 技术路线
_FlashMLA (Kernel for MLA)
DeepEP load balancer
NSA (CSA and HCA are in a way related to it)
EnGram (Embeddings in CPU RAM - will be a game changer, helps model reconstruct the meaning sooner, helps with reasoning)
Recipe to train models for competitive math in DeepSeek Speciale paper
DeepSeek OCR_
tags: 好信源 · → daily
3. 竞争格局:真追平 vs 落后6个月
*topic: 竞争格局/模型发布 · 56 bull vs 13 bear*
🟢 bullish 侧:
- <a id="atom-ea67070e6e6daa44"></a>2026-04-19
X·@tuolaji2024 [老化中] · V4 跑分泄露,各项能力在 benchmark 中达到 SOTA · 模型发布/技术路线
V4跑分泄露,跑分领先,各项能力在benchmark中对比GPT5.3、Claude Opus4.6、Gemini 3.1 Pro Preview来看均达到SOTA
tags: 好信源 · → daily
value: direction=达到SOTA
📷 原图
- <a id="atom-cfa7bfe597405f3f"></a>2026-04-19
X·@teortaxesTex [老化中] · 研究人员流失程度与其他公司对比 · 人事变动/竞争格局
there's been some attrition, but it's not even close to the losses of OpenAI or Meta, to say nothing of XAI. I also don't notice it accelerating.
tags: 好观点 · → daily
📷 原图
- <a id="atom-2a69e752ed5b3883"></a>2026-04-23
X·@emollick [老化中] · 基准测试表现 · 模型发布
Good benchmarks
tags: 好观点 · → daily
- <a id="atom-57fcb44fcee63554"></a>2026-04-23
X·@AIWarper · model 4 release and open source status · 模型发布
Deepseek 4 officially released and open sourced.
tags: 好信源 · → daily
📷 原图
- <a id="atom-ff33d788c50428df"></a>2026-04-23
X·@zizhpan · 发布最强模型V4-Pro · 模型发布
Meet DeepSeek-V4-Pro: our largest and most advanced model to date
tags: 好信源 · → daily
🔴 bearish 侧:
- <a id="atom-389684da7b1fccb8"></a>2026-04-19
X·@theinformation · 下一个模型延迟 · 模型发布
delays to its next model
tags: 好信源 · → daily
- <a id="atom-f24102fc9816525a"></a>2026-04-20
X·@poezhao0605 · 下一代V4模型发布时间多次延期 · 模型发布/产品发布时间
The next-generation V4 model has reportedly slipped from around the Lunar New Year to February, then March, and now late April
tags: 好信源 · → daily
value: date=2026-04-20
- <a id="atom-7005a22dbd18e5ff"></a>2026-04-28
X·@teortaxesTex · 官方参考模型代码 · 模型发布/技术路线 +同日1条
官方参考模型代码输出被损坏
tags: 好信源 · → daily
- <a id="atom-9ccb4eed76ae839f"></a>2026-04-29
X·@_xjdr · 官方推理代码未测试 · 模型发布/技术路线_
官方推理代码未通过主要部署代码测试
tags: 好信源 · → daily
- <a id="atom-2662e617c544a0a7"></a>2026-05-01
X·@teortaxesTex · AA-Omniscience得分表现 · 模型发布/技术路线
It's actually insane how bad DeepSeek does on AA-Omniscience. It actually knows a hell of a lot, but they absolutely have failed or neglected to train for this. This is 12.5% of total score. –10 points for Pro. If not for that one thing, it'd be ranked the best OS model.
tags: 好数字 · → daily
value: qty=-10 points · direction=negative
展开 70 条中性
- <a id="atom-7b1b660956f57bed"></a>2026-04-21
X·@AiBattle_ [老化中] · V3.2-exp 路由到 V4 · 技术路线/模型发布
Deepseek-v3.2-exp was added to the arena about an hour ago and appears to route to V4
tags: 好观点 · → daily
📷 原图
- <a id="atom-7e8fc17a80a61221"></a>2026-04-22
X·@teortaxesTex [老化中] · 模型发布日期策略 · 模型发布
Pessimistic interpretation: since Spud/Chud 5.5 are confirmed as coming out on Thursday, DeepSeek has decided to go for their <24 hours in the limelight. If not for these news, it'd have been 29th as in earlier rumors
tags: 好观点·好思考 · → daily
- <a id="atom-6faa2e24bebd5c3c"></a>2026-04-23
X·@cnfinancewatch [已过期] · V4模型发布传闻时间 · 模型发布
网传DeepSeek V4将于2026年4月下旬正式发布
tags: 好问题 · → daily
value: date=2026年4月下旬
- <a id="atom-e68d3b6561f15a7f"></a>2026-04-23
X·@dhtikna [老化中] · 未来论文引用计划 · 技术路线/模型发布
Rohan, an author of the Google NGrammer, paper emailed Deepseek, who confirmed to Rohan there will be a future paper with an Engram section where they will cite NGrammer (which wasn't cited in the OG Engram paper)
tags: 好信源 · → daily
- <a id="atom-324a191fac79fc00"></a>2026-04-23
X·@eliebakouch · 开源V4 Pro(1.6T)和Flash(292B)模型 · 模型发布
deepseek just open sourced V4 Pro (1.6T) and Flash (292B)
tags: 好信源·好数字 · → daily
value: qty=1.6T in V4 Pro, 292B in Flash
📷 原图
4. 人才争夺:核心研究员离职 vs 流失可控
*topic: 人事变动/竞争格局 · 53 bull vs 40 bear*
🟢 bullish 侧:
- <a id="atom-a00d64e24f85bf6d"></a>2026-04-18
X·@Presidentlin · 追赶差距 · 竞争格局
DeepSeek will come back and be 3 months behind, and not the 6 months behind associated with Chinese models.
tags: 好观点·好数字 · → daily
value: qty=3 months behind · date=2026
- <a id="atom-47f1553c20f2a7e1"></a>2026-04-19
X·@teortaxesTex [老化中] · 竞争策略表现 · 竞争格局
DeepSeek's 'Do nothing, Win' meme only hits harder now
tags: 好观点 · → daily
value: direction=强
- <a id="atom-adfb3de7b19d7e58"></a>2026-04-22
X·@jukan05 · 正在使用 NVIDIA 芯片训练新模型 · 模型发布/竞争格局
DeepSeek upcoming models are still being trained on Nvidia chips
tags: 好信源 · → daily
value: date=2026-04-22
📷 原图
- <a id="atom-1bb879f52623bf68"></a>2026-04-23
X·@ZhihuFrontier [老化中] · 长文本比较表现 · 竞争格局
Outperformed web Gemini, which refused to process the long context
tags: 好观点 · → daily
📷 原图
- <a id="atom-f8bb037fc9ec4f10"></a>2026-04-24
X·@pastaraspberry [老化中] · 模型发布优势 · 竞争格局/技术路线
Really hope that Wenfeng will continue to mog you and beat you smug "our models are so dangerous to release, muh distillation"
tags: 好观点 · → daily
🔴 bearish 侧:
- <a id="atom-f05c653fa1db34d1"></a>2026-04-15
X·@Presidentlin [老化中] · 被华为首发 · 竞争格局/地缘政治
The day that DeepSeek comes out on Huawei first, that is a horrible outcome for our nation.
tags: 好观点 · → daily
📷 原图
- <a id="atom-8982151e298d9407"></a>2026-04-18
X·@FundaAI [老化中] · 未融资导致薪酬竞争力不足 · 竞争格局/人事变动
DeepSeek's main issue is that it hasn't done any fundraising before, so its compensation packages can't keep up with ByteDance and other frontier labs. As a result, a significant number of its core researchers have been poached.
tags: 好观点 · → daily
- <a id="atom-a57a12f73a180147"></a>2026-04-19
X·@theinformation · 研究人员离职 · 人事变动
The move comes after researcher departures
tags: 好信源 · → daily
- <a id="atom-06ad281e559ac515"></a>2026-04-20
X·@poezhao0605 · 核心研究员离职 · 人事变动
in the same period, at least 5 core researchers have left for competitors
tags: 好数字·好信源 · → daily
value: qty=5 · date=同期
- <a id="atom-f64ff96fa5af636d"></a>2026-04-24
X·@poezhao0605 · researcher departures · 人才流失/人事变动
At least 5 senior researchers have left since late 2025. Guo Daya, a key architect behind R1's GRPO algorithm with 37,000+ citations, joined ByteDance in April.
tags: 好数字·好信源 · → daily
value: qty=at least 5
展开 44 条中性
- <a id="atom-1b431c855971d537"></a>2026-04-15
X·@teortaxesTex · 不会在华为或其他国内平台训练V4 · 技术路线/竞争格局
DeepSeek V4 won't be trained on Huawei or any other domestic platform
tags: 好观点·好思考 · → daily
📷 原图
- <a id="atom-39263d28c42cbb78"></a>2026-04-19
X·@teortaxesTex · 核心研究人员统计 · 人事变动
Ruan Chong is present on 18 papers, then Daya Guo (11), Bingxuan Wang (9), Fuli Luo (8), Haoran Wei (3). Those are big figures. But Zhenda Xie (16), Damai Dai (14), Yu Wu (14), Qihao Zhu (10), Runxin Xu (9)… etc. are still there as of V3.2
tags: 好数字 · → daily
📷 原图
- <a id="atom-263a44624d9f50aa"></a>2026-04-25
X·@teortaxesTex · 招聘了 Anyi Xu · 人事变动
Anyi Xu has come onboard somewhere between V3.2 and V4 it seems
tags: 好观点 · → daily
value: date=V3.2 和 V4 之间
📷 原图
- <a id="atom-803f64f14060b4d2"></a>2026-04-26
X·@teortaxesTex · NSA贡献者身份 · 人事变动/技术路线
first author of NSA was an intern at DeepSeek
tags: 好信源 · → daily
- <a id="atom-728fadd3a1a62af3"></a>2026-04-28
X·@teortaxesTex [老化中] · 模型特性 · 模型能力/竞争格局
DeepSeek makes the most hallucinatory smart models, also horrible at pushing back on the user's bullshit
tags: 好思考·好观点 · → daily
value: direction=最严重
📷 原图
5. 产能与算力:华为昇腾破局 vs 算力瓶颈
*topic: 供给产能/数据中心算力/价格动态 · 36 bull vs 32 bear*
🟢 bullish 侧:
- <a id="atom-9061f3b542240c82"></a>2026-04-16
X·@jiqizhixin · Mega MoE 性能效果 · 数据中心算力
The result is a shift from fragmented execution to a continuous pipeline: higher GPU utilization, less idle time, and much better scaling in multi-GPU MoE workloads.
tags: 好信源·好思考 · → daily
📷 原图
- <a id="atom-ab1b6fe437b3f71c"></a>2026-04-23
X·@TensorSlay · 芯片间带宽饱和度对比 · 价格动态 +同日1条
inter-chip bandwidth saturation drops from 92 % (H800 mesh) to 73 % (910C torus)
tags: 好数字 · → daily
value: qty=92% vs 73% · date= · direction=下降
- <a id="atom-26b576aca334509e"></a>2026-04-23
X·@tuolaji2024 · Pro 价格预计下调 · 价格动态/产品发布
预计下半年昇腾950超节点批量上市后,pro 的价格会大幅下调。
tags: 好思考 · → daily
[图: DeepSeek API 新模型 V4-Pro 与 V4-Flash 的定价及参数表格 — deepseek-v4-pro输入(缓存未命中): 12元; deepseek-v4-pro输出: 24元; deepseek-v4-flash输入(缓存未命中): 1元; deepseek-v4-flash输出: 2元; 上下文长度: 1M]
📷 原图
- <a id="atom-4769558e4f2f0cf2"></a>2026-04-24
X·@bookwormengr · KV cache 削减幅度 · 技术路线/数据中心算力
Cutting down KV cache by 90% without degrading performance (reducing HBM demand)
tags: 好数字·好信源 · → daily
value: qty=90% · direction=reducing
- <a id="atom-d56754142b1761fd"></a>2026-04-24
X·@dhtikna · next gen cluster access and mastery timeline · 技术路线/供给产能
Things may change 2027 when Huawei next gen clusters come online and deepseek gets a hang of them
tags: 好观点 · → daily
value: date=2027
🔴 bearish 侧:
- <a id="atom-3ddd0915252f46c7"></a>2026-04-20
X·@teortaxesTex [老旧] · 算力集群获取 · 供给产能
it's relatively easy to collect an OOM bigger Nvidia cluster than what they have
tags: 好观点 · → daily
- <a id="atom-0a85a7ac88c1a079"></a>2026-04-23
X·@tuolaji2024 · Pro 服务吞吐受限于高端算力 · 供给产能/算力瓶颈
受限于高端算力,目前Pro的服务吞吐十分有限,会大幅下调。
tags: 好观点 · → daily
[图: DeepSeek API 新模型 V4-Pro 与 V4-Flash 的定价及参数表格 — deepseek-v4-pro输入(缓存未命中): 12元; deepseek-v4-pro输出: 24元; deepseek-v4-flash输入(缓存未命中): 1元; deepseek-v4-flash输出: 2元; 上下文长度: 1M]
📷 原图
- <a id="atom-5cb8f519cf3b6739"></a>2026-04-23
X·@poezhao0605 · V4-Pro 价格预期 · 价格动态/数据中心算力
Prices will drop significantly once Huawei Ascend 950 super nodes ship at scale in H2.
tags: 好观点 · → daily
value: date=H2 2026
[图: DeepSeek-v4系列模型API访问价格与上下文长度对照表 — deepseek-v4-pro输入(缓存未命中): 12元; deepseek-v4-pro输出: 24元; deepseek-v4-flash输入(缓存未命中): 1元; deepseek-v4-flash输出: 2元; 上下文长度: 1M]
📷 原图
- <a id="atom-3e95e95ec9792c3a"></a>2026-04-24
X·@pstAsiatech · V4-pro吞吐量限制原因 · 供给产能/数据中心算力 +同日1条
the throughput of V4-pro was currently limited by a shortfall in computational supply
tags: 好观点 · → daily
- <a id="atom-bbdc81243af230e2"></a>2026-04-26
X·@poezhao0605 · 价格下降预期 · 价格动态/供给产能
DeepSeek confirmed again: prices will drop further once Huawei Ascend 950 nodes ship at scale in H2.
tags: 好观点 · → daily
value: date=H2
📷 原图
展开 46 条中性
- <a id="atom-10f9535df47c4576"></a>2026-04-15
X·@Dorialexander [老化中] · 利用 DSA 降低长上下文价格 · 价格动态
DSA allowed DeepSeek to cut prices agressively on long context
tags: 好观点 · → daily
📷 原图
- <a id="atom-a02ad19b7ce13152"></a>2026-04-16
X·@jiqizhixin · Mega MoE 架构特点 · 技术路线/数据中心算力
Instead of running MoE as a chain of separate steps (dispatch → MLP → combine), Mega MoE fuses everything into a single mega-kernel. Even more importantly, it overlaps NVLink communication with Tensor Core computation, reducing the classic “compute–wait–transfer” bottleneck.
tags: 好信源 · → daily
📷 原图
- <a id="atom-7fd7963fb4880901"></a>2026-04-25
X·@teortaxesTex · 下一步行动-内蒙古数据中心 · 供给产能/资本开支
complete their Inner Mongolia datacenters, proceed to train a simplified, larger V5 with 100T+ tokens
tags: 好观点·好数字 · → daily
- <a id="atom-e520a4462fdf8f01"></a>2026-04-28
X·@teortaxesTex · Pro 高端算力有限导致服务吞吐受限 · 供给产能
Due to limitations in high-end computing power, the service throughput of Pro is currently very limited
tags: 好信源 · → daily
value: date=当前
[图: DeepSeek V4-Pro 与 V4-Flash API 价格及规格表 — deepseek-v4-pro输入(缓存未命中): 12元; deepseek-v4-pro输出: 24元; deepseek-v4-flash输入(缓存未命中): 1元; deepseek-v4-flash输出: 2元; 最大上下文长度: 1M]
📷 原图
- <a id="atom-42d5151f0d158784"></a>2026-04-27
X·@teortaxesTex [老化中] · 当前训练硬件组合 · 供给产能
I suspect that they currently generate rollouts with Huawei and train exclusively on NVDA (and their Hopper stock at that)
tags: 好思考 · → daily
📷 原图
6. 开源模式:重塑行业叙事 vs 变现死胡同
*topic: 商业模式/定价权 · 18 bull vs 5 bear*
🟢 bullish 侧:
- <a id="atom-293f0f37427f4947"></a>2026-04-27
X·@zerohedge · aggressively pitching low-priced plans for just-released flagship model, intensifying competition in Chinese AI industry · 产品发布/定价权
DeepSeek is aggressively pitching low-priced-plans for its just-released flagship model, intensifying competition across a Chinese artificial intelligence industry trying to take on Silicon Valley’s best. BBG
tags: 好信源 · → daily
- <a id="atom-c7b4a2c2fd65c805"></a>2026-05-01
X·@Presidentlin · 可能按等待时长给予不同折扣 · 定价权 +同日1条
Maybe they will even do something crazy and give different discounts for the wait time.
tags: 好思考 · → daily
- <a id="atom-c070bfe896396b2e"></a>2026-05-01
X·@teortaxesTex [老化中] · 90%缓存削减可使成本进一步降低 · 定价权/供给产能
with DS 90% cache cut, would be cheaper still
tags: 好观点 · → daily
- <a id="atom-82cd56cf471d7f21"></a>2026-05-04
X·@teortaxesTex · cache价格不会被调高 · 定价权
the cache price is not going to change, it's driven by fundamental low cost
tags: 好思考 · → daily
value: direction=unchanged
- <a id="atom-4c08a4aadede6613"></a>2026-05-10
X·@teortaxesTex · 定价能力 · 定价权
they'll be able to raise prices with impunity
tags: 好观点 · → daily
🔴 bearish 侧:
- <a id="atom-8c3ed927ccd21c5b"></a>2026-04-17
X·@teortaxesTex [老化中] · 开源商业模式 · 商业模式
bit hard to raise money if your whole pitch is you're going to open source
tags: 好观点 · → daily
- <a id="atom-56a58fcc518e5a76"></a>2026-05-23
X·@dhtikna · 定价行为 · 定价权
Dont say chinese labs, say deepseek. Other chinese labs will charge what they can get away with
tags: 好观点 · → daily
value: qty=aggressive pricing
- <a id="atom-0c952d9a25b05df6"></a>2026-05-25
X·@bfmbusiness · 定价策略描述 · 定价权/竞争格局
"C'est une stratégie de dumping : il distribue massivement, quasi gratuitement, leurs modèles, et à côté ils font en sorte que tout le monde aille sur Alibaba car le token est très bas"
tags: 好观点 · → daily
- <a id="atom-92777d93e5fd28c3"></a>2026-05-29
X·@hsu_steve · 对 US AI labs 的 token 定价压力 · 定价权
Token pricing pressure on US AI labs from DeepSeek
tags: 好观点 · → daily
📷 原图
- <a id="atom-1b052494b8819984"></a>2026-06-27
X·@Pier4r · API 定价 · 定价权/价格动态
API prices are profitable but 1/2
tags: 好观点·好思考 · → daily
value: qty=亏损 · currency=USD · date=2026
展开 10 条中性
- <a id="atom-96c1deae0b7c4e6e"></a>2026-04-17
X·@swyx [老化中] · 开源哲学立场 · 商业模式
they're probably not as philosophically purist on open source as like an AI2 or a huggingface
tags: 好观点 · → daily
- <a id="atom-2a38780eb3359253"></a>2026-04-27
X·@teortaxesTex [老化中] · 降价策略迫使华为加速 · 定价权
I think DeepSeek is rushing them with these price slashes, after announcing it in H2'26 at release
tags: 好思考 · → daily
📷 原图
- <a id="atom-fe9dccd230f0a139"></a>2026-05-04
X·@mcndjxlefnd [老化中] · 定价促销截止日 · 定价权
Isn't this just a limited time offer (till May 1st)?
tags: 好问题 · → daily
value: date=2026-05-01 · direction=before
- <a id="atom-ef9089eae2b41927"></a>2026-05-14
X·@ryaneshea · 最佳AI模型——成本维度 · 定价权
The answer changes based on what you optimize for: Cost → DeepSeek, MiniMax
tags: 好观点 · → daily
📷 原图
- <a id="atom-380da2e1a14df3db"></a>2026-05-15
X·@covrovski · pricing action · 定价权
Deepseek pretty much the only one lowering prices?
tags: 好问题 · → daily
value: direction=decreasing
🔗 因果传导 (causal map)
*DeepSeek 在产业链上的传导关系. 边是群里/卖方陈述的因果 (非 AI 推断), 数字 = 几条 atom 支撑. 点 atom 溯源.*
↓ 下游·近期陈述 (7)
- 🟢 zAI · 替代 · 2 次陈述
When DeepSeek figured out MLA that saves KV Cache by 93%, many major labs like ZAI and MoonShot adopted it.
- 🔴 American AI models · 替代 · 1 次陈述
move to Deep seek models from American home grown models flag that AI is China shock on steroids
- 🔴 AI stocks · 成本传导 · 1 次陈述
The drop in prices will have to be made up somewhere (via faster adoption), otherwise doesn’t bode well for stocks (but
- 🟢 MSFT · 成本传导 · 1 次陈述
MSFT is considering leveraging DeepSeek, a cheaper Chinese model, to support Copilot Cowork.
- 🟢 MoonShot · 替代 · 1 次陈述
When DeepSeek figured out MLA that saves KV Cache by 93%, many major labs like ZAI and MoonShot adopted it.
- ⚪ fundraising · 传导 · 1 次陈述
Anthropic’s Mythos preview reportedly pushed DeepSeek into a $7.4B fundraising - because they could not compete with Myt
- 🟢 存储标的 · 需求拉动 · 1 次陈述
DeepSeek 当前的 SSD 配置和利用率已大幅提升,直接印证了存储标的的上涨逻辑
↑ 上游·近期陈述 (9)
- 🟢 Anthropic · 竞争 · 1 次陈述
Anthropic’s Mythos spurred Deepseek fundraise
- 🟢 distillation · 传导 · 1 次陈述
Yes, DS has good pretraining, but the reason they’re able to compete on the price-performance frontier is distillation.
- ⚪ Kimi · 产能依赖 · 1 次陈述
Kimi and Zhipu's flagships run on DeepSeek's older designs.
- ⚪ Zhipu · 产能依赖 · 1 次陈述
Kimi and Zhipu's flagships run on DeepSeek's older designs.
- 🔴 人才流失 · 传导 · 1 次陈述
We know they lost good people. The hash-based router, heuristic kitchen sink approach to stability, late release, and re
- 🔴 lack of war chest · 竞争 · 1 次陈述
Liang reportedly concluded DeepSeek could not compete at that level without a much larger war chest.
- 🔴 Anthropic's Mythos preview · 竞争 · 1 次陈述
Anthropic’s Mythos preview reportedly pushed DeepSeek into a $7.4B fundraising - because they could not compete with Myt
- 🔴 Anthropic Mythos preview · 竞争 · 1 次陈述
Anthropic’s Mythos preview reportedly pushed DeepSeek into a $7.4B fundraising
📏 估计带 (5)
*语料内对同一量的估计区间 (≥2 个估计才成带). 不判断谁对, 只陈列.*
- 估值倍数 未标期限 ⚠
混合口径: $27,390 – $12B · 3 个估计 · 最新 2026-06-29 $12B (X·cred3)
- 份额 未标期限 ⚠
混合口径: 17% – 50% · 2 个估计 · 最新 2026-06-27 50% (X·cred3)
📊 程序化变化 (近 7 天)
- 新增 1 atoms
- 本周新 topic: AI投资/市场情绪
- 已过期 forecast: 2 条
展开过期列表
[已过期] 2025H2 · 价格预期
prices will drop significantly in the second half of the year once Huawei's Ascend 950PR supernodes ship at scale
[已过期] 2025-06 · 模型更新计划
DEEPSEEK AIMS TO RELEASE UPDATE TO V4 MODEL IN JUNE:INFORMATION
🟦 客观事实 (facts) (30)
- <a id="atom-08baaa97ac6cd624"></a>🟦 2026-06-29
X·@theinformation ⭐ · 本轮融资规模 · 融资
DeepSeek raise $7.4B
tags: 好数字·好信源 · → daily
value: qty=$7.4B · date=2026-06-29
- <a id="atom-f4ef3292cd7c0aa5"></a>🟦 2026-06-27
X·@theinformation ⭐ · 融资 · 融资
Exclusive: DeepSeek raised $7.4 billion after Anthropic’s Mythos preview convinced CEO Liang Wenfeng that China’s top AI lab needed far more money to compete.
tags: 好数字·好信源 · → daily
value: qty=$7.4B
- <a id="atom-f00866f03bec614a"></a>🟦 2026-06-16
X·@dmjk001 ⭐ · 本轮投资额 · 融资
该国家基金本轮投资额为 10 亿元人民币。
tags: 好数字·好信源 · → daily
value: qty=10亿元人民币
- <a id="atom-a0cd5a2b1a2c6a4f"></a>🟦 2026-06-16
X·@dmjk001 ⭐ · 首轮融资募资金额 · 融资
深度求索(DeepSeek)已完成首轮融资,以非传统的交易架构筹集逾 500 亿元人民币(约合 74 亿美元)。
tags: 好数字·好信源 · → daily
value: qty=500 亿元人民币 · date=
📷 原图
- <a id="atom-c279f586dc2721e4"></a>🟦 2026-06-16
X·@dmjk001 ⭐ · 首轮融资估值 · 融资/估值
本轮融资对这家大语言模型开发商的估值超过500亿美元。
tags: 好数字·好信源 · → daily
value: qty=超过500亿美元 · date=
📷 原图
- <a id="atom-47cce971c520353d"></a>🟦 2026-06-16
X·@dmjk001 ⭐ · 投资者股份锁定期 · 融资
该架构对所有投资者所持股份设定了五年锁定期,期间不得出售股权。
tags: 好信源·好数字 · → daily
value: qty=五年 · date=
📷 原图
- <a id="atom-ece9f78db91e7c17"></a>🟦 2026-06-16
X·@theinformation ⭐ · 完成新一轮融资 · 融资
DeepSeek just closed a major $7.4B new funding round.
tags: 好数字·好信源 · → daily
value: qty=$7.4B
- <a id="atom-942959e1e7b9ecdc"></a>🟦 2026-06-07
X·@poezhao0605 ⭐ · 融资规模 · 融资
DeepSeek is raising $7.4 billion at up to $59 billion.
tags: 好数字·好信源 · → daily
value: qty=$7.4 billion · date=2026-06-07
📷 原图
- <a id="atom-eb28633147a2a08d"></a>🟦 2026-06-07
X·@poezhao0605 ⭐ · 估值 · 估值
DeepSeek is raising $7.4 billion at up to $59 billion.
tags: 好数字·好信源 · → daily
value: qty=$59 billion · date=2026-06-07
📷 原图
- <a id="atom-5b2c607bccfbce73"></a>🟦 2026-06-03
X·@dmjk001 ⭐ · 首轮融资拟筹金额 · 融资
DeepSeek 首轮融资拟筹 70 亿美元
tags: 好数字·好信源 · → daily
value: qty=70亿美元
📷 原图
- <a id="atom-eaa0298febe49e1d"></a>🟦 2026-06-03
X·@Techmeme ⭐ · 第一轮融资规模 · 融资
Sources: DeepSeek is set to raise ~$7.4B in its first funding round from investors including Tencent and CATL at a valuation of between ~$52B and ~$59B (Reuters)
tags: 好数字·好信源 · → daily
value: qty=~$7.4B · date=2026-06-03
- <a id="atom-25f185c1c2c618c6"></a>🟦 2026-06-03
X·@yicaichina ⭐ · 第一轮融资金额 · 融资
DeepSeek is set to bag about CNY50 billion (USD7.4 billion) in its first funding round
tags: 好数字·好信源 · → daily
value: qty=CNY50B (USD7.4B) · date=第一轮
📷 原图
- <a id="atom-a27adabaef28879f"></a>🟦 2026-06-03
newsletter·Reuters ⭐ · 首轮融资 · 融资/估值
DeepSeek slated to raise $7 billion in maiden funding round - Reuters; post-money valuation ¥350B-¥400B ($52B-$59B).
tags: 好数字·好信源 · → daily
value: qty=$7 billion, post-money valuation ¥350B-¥400B ($52B-$59B)
- <a id="atom-ff00a88500b75a23"></a>🟦 2026-05-28
X·@zerohedge ⭐ · token pricing cut · 价格动态
DeepSeek reportedly cut token pricing by 75%
tags: 好数字·好信源 · → daily
value: qty=75% · direction=down
- <a id="atom-68a508a36bd6cc99"></a>🟦 2026-05-27
X·@tuolaji2024 ⭐ · 投前估值 · 估值/融资
国家大基金领投DeepSeek,投前估值450亿美元。
tags: 好数字·好信源 · → daily
value: qty=450亿美元
- <a id="atom-5e330e2372470be3"></a>🟦 2026-05-26
X·@pstAsiatech ⭐ · 融资估值 · 融资/估值
AI-model maker DeepSeek, long known for avoiding outside fundraising, began sounding out investors to raise capital at a valuation of about 300 billion yuan ($44 billion).
tags: 好数字·好信源 · → daily
value: qty=300 billion yuan ($44 billion) · date=2026-05-26
- <a id="atom-3d9665b753b27104"></a>🟦 2026-05-24
X·@rohanpaul_ai ⭐ · V4-Pro price cut magnitude · 价格动态
Reuters has already reported a permanent 75% DeepSeek V4-Pro price cut
tags: 好数字·好信源 · → daily
value: qty=75%
📷 原图
- <a id="atom-790c8bc1abb67bfe"></a>🟦 2026-05-22
X·@dmjk001 ⭐ · 正在进行融资轮规模 · 融资
正在进行 700 亿元人民币(100 亿美元)融资轮
tags: 好数字·好信源 · → daily
value: qty=700亿元人民币(100亿美元)
📷 原图
- <a id="atom-9db0cd7c8ed7aca2"></a>🟦 2026-05-22
X·@Techmeme ⭐ · 本轮融资规模 · 融资
DeepSeek execs told potential investors in its ongoing $10B round that it will prioritize groundbreaking AI research over short-term commercialization
tags: 好数字·好信源 · → daily
value: qty=$10B · date=ongoing
- <a id="atom-a6d7b5a1245fad80"></a>🟦 2026-05-22
X·@dmjk001 ⭐ · 首轮融资目标金额 · 融资
深度求索的首轮融资旨在筹集约 500 亿元人民币(约 73.5 亿美元),最早可能在下个月完成
tags: 好数字·好信源 · → daily
value: qty=500亿元(约73.5亿美元) · date=2026-06
📷 原图
- <a id="atom-f0441a5ec8308e13"></a>🟦 2026-05-22
X·@dmjk001 ⭐ · 融资后估值 · 融资
交易完成后,深度求索的估值可能超过 3500 亿元人民币(约 514 亿美元)
tags: 好数字·好信源 · → daily
value: qty=超过3500亿元(约514亿美元) · date=2026-06
📷 原图
- <a id="atom-0759d410b0360ca1"></a>🟦 2026-05-22
X·@scaling01 ⭐ · API价格调整状态 · 价格动态
DeepSeek-V4-Pro 模型 API 价格将于2026年5月31日结束2.5折优惠活动后,正式调整为原定价的1/4。调整后每百万token输入3元人民币,输出6元人民币
tags: 好数字·好信源 · → daily
value: date=2026-05-31 · direction=down
- <a id="atom-99dd49db1f461475"></a>🟦 2026-05-08
X·@jukan05 ⭐ · 融资目标金额 · 融资
DeepSeek 融资目标高达 500 亿人民币 (约 73.5 亿美元)
tags: 好数字·好信源 · → daily
value: qty=RMB 50 billion ($7.35 billion) · direction=seeking
- <a id="atom-3bc600f562d102db"></a>🟦 2026-05-08
X·@jukan05 ⭐ · 本轮融资历史地位 · 融资
若完成,将成为中国AI公司历史上最大单笔融资
tags: 好数字·好信源 · → daily
value: qty=最大单笔融资
- <a id="atom-cf2289c8ec5f6189"></a>🟦 2026-05-08
X·@jukan05 ⭐ · V4.1 发布时间 · 模型发布/产品发布
公司计划于 6 月发布 V4.1 版本
tags: 好数字·好信源 · → daily
value: qty=June 2026
- <a id="atom-9562ab9fb3a024be"></a>🟦 2026-05-08
X·@dmjk001 ⭐ [老化中] · 本轮融资额预期 · 融资
The Information:DeepSeek希望本轮融资额高达 500 亿元人民币(约合 73.5 亿美元)
tags: 好数字·好信源 · → daily
value: qty=500 亿元人民币(约合 73.5 亿美元)
📷 原图
- <a id="atom-9b1baa947fed9eb3"></a>🟦 2026-05-08
X·@dmjk001 ⭐ [老化中] · 融资后估值预期 · 融资
预期估值已大幅攀升 ,从最初超过 100 亿美元的预期,升至本轮融资后可能超过 3500 亿元人民币(约合 515 亿美元),包含所筹资金。
tags: 好数字·好信源 · → daily
value: qty=超过 3500 亿元人民币(约合 515 亿美元)
📷 原图
- <a id="atom-e2254b148b7c534f"></a>🟦 2026-05-08
X·@teortaxesTex ⭐ · 本轮融资规模 · 融资
DeepSeek is targeting a fundraising size of up to RMB 50 billion, or approximately $7.35 billion, in this round. If completed, it could mark the largest single fundraising round in the history of Chi
tags: 好数字·好信源 · → daily
value: qty=RMB 50 billion / $7.35 billion · date=2026-05-08
- <a id="atom-11516873f034212b"></a>🟦 2026-05-06
X·@dmjk001 ⭐ [老化中] · 估值 · 融资/估值
DeepSeek 估值逼近 450 亿美元
tags: 好数字·好信源 · → daily
value: qty=约450亿美元 · date=首轮融资
📷 原图
- <a id="atom-f5fa4d53b0790867"></a>🟦 2026-05-06
X·@jukan05 ⭐ · first fundraising round valuation · 融资
FT: DeepSeek is in talks for its first fundraising round at a valuation of around $45 billion.
tags: 好数字·好信源 · → daily
value: qty=~$45 billion · date=2026
📷 原图
🟥 多头 takes (bullish) (30)
- <a id="atom-ef89d0dea3aa28c3"></a>🟥 2026-06-27
X·@scaling01 · is not profitable at $0.87 · 定价权/盈利能力
DeepSeek is not profitable at $0.87
tags: 好观点·好信源 · → daily
- <a id="atom-659aa2b5a1b09ce1"></a>🟥 2026-05-06
X·@AndrewCurran_ · 估值 · 估值
DeepSeek could be valued at up to $50 billion after its first funding round
tags: 好数字·好观点 · → daily
value: qty=$50 billion · date=first funding round
📷 原图
- <a id="atom-dacbb5c48c83efc3"></a>🟥 2026-06-30
卖·MS Tom Wigg · 对开源模型叙事的影响 · 竞争/叙事
This hike potentially challenges the very narrative set off by DeepSeek's core offering.
tags: 好观点·好思考 · → daily
- <a id="atom-23d95771e6b8ab96"></a>🟥 2026-06-25
X·@teortaxesTex · 团队规模预期 · 团队扩张
I expect them to exceed 1000 members soon
tags: 好数字·好观点 · → daily
value: qty=超过1000人 · date=soon
📷 原图
- <a id="atom-dc5861b44af23663"></a>🟥 2026-06-18
X·@teortaxesTex · H100等效算力预期 · 供给产能
I expect them to have at least 100K H100-equivalents by EoY
tags: 好数字·好观点 · → daily
value: qty=100K H100-equivalents · date=2026 EoY
- <a id="atom-0db3398cef4d0b09"></a>🟥 2026-05-17
X·@yujianfeng55 · token 成本 · 定价权
烧了两千万token,花了一毛人民币
tags: 好数字·好观点 · → daily
value: qty=2000万 token 0.1元
📷 原图
- <a id="atom-a00d64e24f85bf6d"></a>🟥 2026-04-18
X·@Presidentlin · 追赶差距 · 竞争格局
DeepSeek will come back and be 3 months behind, and not the 6 months behind associated with Chinese models.
tags: 好观点·好数字 · → daily
value: qty=3 months behind · date=2026
- <a id="atom-3e8abed8d7b5edea"></a>🟥 2026-04-29
群 · SSD 配置升级实证 · SSD配置升级/存储需求
DeepSeek 当前的 SSD 配置和利用率已大幅提升,直接印证了存储标的的上涨逻辑
tags: 好观点·好信源 · → daily
- <a id="atom-53fd990e6ffc7d55"></a>🟥 2026-06-30
X·@teortaxesTex · built a system enabling zany experiments unparalleled · 技术路线/研发效率
DS has built a system that affords them very zany experiments, it's unparalleled
tags: 好观点·好信源 · → daily
value: direction=unique
📷 原图
- <a id="atom-487c9762c19934bf"></a>🟥 2026-06-16
X·@scaling01 ⭐ · 未来模型参数规模预期 · 模型参数/技术路线
also try to imagine what a 1.6T DeepSeek model will do
tags: 好思考·好问题 · → daily
value: qty=1.6T
📷 原图
- <a id="atom-0004878d1d63d697"></a>🟥 2026-05-09
X·@ZhihuFrontier [老化中] · KV Cache optimization groundbreaking · 推理效率/技术领先
After running 20,000 controlled LLM inference experiments over 3 straight days, I’ve fully uncovered how groundbreaking DeepSeek’s KV Cache optimization really is.
tags: 好观点·好信源 · → daily
[图: 大语言模型API Token使用量与缓存命中情况的数据面板图表 — 总Token数: 50,949,110; 2026-04-28单日Token数: 21,213,830; 输入(命中缓存): 20,526,208; 输入(未命中缓存): 443,048; 输出: 244,574]
📷 原图
- <a id="atom-b9da019bfa51dc14"></a>🟥 2026-05-09
X·@ZhihuFrontier [老化中] · KV Cache hit rate impact on costs · 推理成本/缓存命中率
its exceptional cache hit rate drastically cuts token & inference costs.
tags: 好观点·好信源 · → daily
[图: 大语言模型API Token使用量与缓存命中情况的数据面板图表 — 总Token数: 50,949,110; 2026-04-28单日Token数: 21,213,830; 输入(命中缓存): 20,526,208; 输入(未命中缓存): 443,048; 输出: 244,574]
📷 原图
- <a id="atom-e714ac7de9bbd656"></a>🟥 2026-04-28
X·@teortaxesTex [老旧] · TileLang adoption · 技术路线/竞争格局
Basically TileLang is now part of Official DeepSeek Stack.
tags: 好观点·好信源 · → daily
value: qty=yes · date=current
📷 原图
- <a id="atom-5221187c2f17fc1f"></a>🟥 2026-04-26
X·@teortaxesTex [老旧] · CSA+HCA项目成效 · 技术路线
CSA+HCA basically implements its ideas better
tags: 好观点·好信源 · → daily
- <a id="atom-d829af480aaa4bf8"></a>🟥 2026-04-30
X·@teortaxesTex · V4.2 技术路线预测 · 技术路线
they'll go like this: more V4Pro post-train via OPD => new very strong V4.1 generates trillions of tokens of synthetic data with knowledge base grounding => continued pretrain > OPD > V4.2
tags: 好观点·好思考 · → daily
- <a id="atom-d924f39b2c3a6bf6"></a>🟥 2026-06-30
X·@teortaxesTex · V4 vs GLM 5.2 RL cost ratio · 技术路线/成本优势
how much cheaper you can do RL on V4 vs GLM 5.2. It might be a >7x difference depending on the nature of the context
tags: 好观点·好思考 · → daily
value: qty=7x · comparison=cheaper per unit of compute
- <a id="atom-3ac2962a80b40bf2"></a>🟥 2026-06-30
X·@teortaxesTex · V4 base model capability makes product momentum loss affordable in medium term · 产品发布/技术路线
Add a bigger more capable base model, and they really can afford to lose product momentum in the medium term
tags: 好观点·好思考 · → daily
value: direction=better
- <a id="atom-95bce0299b6768e1"></a>🟥 2026-06-30
X·@teortaxesTex · strategic coherence for AGI trajectory · 技术路线/竞争格局
but the big picture is strategically coherent and should afford them a better trajectory towards AGI
tags: 好观点·好思考 · → daily
value: direction=positive
- <a id="atom-9b8bfc6f24e20202"></a>🟥 2026-06-27
X·@teortaxesTex · 2026年招聘立场 · 招聘力度/AGI路线
DeepSeek job postings this season go almost unbelievably hard. They treat this as a final sprint to AGI.
tags: 好观点·好思考 · → daily
value: direction=aggressive
📷 原图
- <a id="atom-fd6edbd692129723"></a>🟥 2026-06-27
X·@brickroad7 · 开放模型全球采用趋势 · 技术路线/地缘政治
Chinese open models become the global default by 2030
tags: 好观点·好思考 · → daily
value: date=2030
- <a id="atom-436723b5d589012e"></a>🟥 2026-06-24
X·@teortaxesTex · 可能重新夺回行业领先地位 · 竞争格局/技术路线
if they don't expand capacity soon DeepSeek will almost definitely reclaim its crown among AGI startups
tags: 好观点·好思考 · → daily
- <a id="atom-c473d797a40ca349"></a>🟥 2026-06-17
X·@teortaxesTex · open-source model release advantage · 模型发布
The Chinese sometimes have vastly superior implementations (eg nobody can beat DeepSeek at serving V4 with full cache support).
tags: 好观点·好思考 · → daily
- <a id="atom-722c883daae3dcc4"></a>🟥 2026-06-15
X·@Alisvolatprop12 · 用户评价排名 · 竞争格局/技术路线
KIMI >= DS/GLM >>> Minimax 정도 입니다.
tags: 好观点·好思考 · → daily
- <a id="atom-a6584de7b8697f62"></a>🟥 2026-06-07
X·@poezhao0605 · 估值变化趋势 · 估值
DeepSeek is no longer being valued as a model lab.
tags: 好观点·好思考 · → daily
📷 原图
- <a id="atom-4d785d3aa3568471"></a>🟥 2026-05-22
X·@deliprao · 成本削减幅度 · 成本优势
deepseek seems to be cutting costs by an order of magnitude than their competition.
tags: 好观点·好思考 · → daily
value: qty=一个数量级
📷 原图
- <a id="atom-54894a53c17507a8"></a>🟥 2026-05-14
X·@teortaxesTex · improvement capability with higher pass@k due to less RL pressure · 技术路线
We know that current generation DS and Kimi can improve a lot with higher k, because they've had less RL pressure.
tags: 好观点·好思考 · → daily
- <a id="atom-d11fd873c14a0d1c"></a>🟥 2026-05-09
X·@teortaxesTex · extreme multi-turn performance potential · 模型性能
If they train it some more, it'll be monstrous for extreme multi-turn
tags: 好观点·好思考 · → daily
value: direction=it'll be monstrous for extreme multi-turn
📷 原图
- <a id="atom-556d49be008c2962"></a>🟥 2026-05-08
X·@cryptopunk7213 · 模型发布预期 · 模型发布
deepseek v4.1 is expected to release soon
tags: 好数字 · → daily
📷 原图
- <a id="atom-8a7efa59f22acdc6"></a>🟥 2026-05-06
X·@teortaxesTex [老化中] · 可通过补贴算力轻松实现AGI商品化 · 资本开支/竞争格局
AGI must be commoditized and this is easily done by subsidizing compute for DeepSeek
tags: 好观点·好思考 · → daily
📷 原图
- <a id="atom-b25cabf01fe638bc"></a>🟥 2026-05-03
X·@LudditeHacker [老化中] · 用户口碑扩散观察 · 订单需求/用户采纳
Deepseek seems to have a slow burn moment rather than the hype of last time. I talked to someone not connected to the discourse and he said he switched to deepseek completely for everything. Strong signal for me
tags: 好观点·好思考 · → daily
🟥 空头 takes (bearish) (30)
- <a id="atom-749eb1a2daad4cae"></a>🟥 2026-05-27
X·@notjazii · v4 Pro定价变化 · 价格动态/竞争格局
deepseek made its v4 Pro model cheaper for good
tags: 好数字·好信源 · → daily
value: qty=降到$0.01 · date=2026-05-27
[图: 小米 mimo-v2.5-pro 与 DeepSeek deepseek-v4-pro 的 API 价格对比表格 — mimo-v2.5-pro常规输入缓存命中价格: $0.20; deepseek-v4-pro常规输入缓存命中价格: $0.01; mimo-v2.5-pro常规输出价格: $3.00; deepseek-v4-pro常规输出价格:
📷 原图
- <a id="atom-657fc364986430d1"></a>🟥 2026-05-26
X·@Krubbled · 计算量差距 · 竞争格局/资本开支
still 4x less then composer 2.5. it will take a while for deepseek to 4x their compute
tags: 好数字·好思考 · → daily
value: qty=4x less then composer 2.5 · direction=still 4x less
- <a id="atom-10c8965d2e7a88fa"></a>🟥 2026-05-06
X·@teortaxesTex · V4-Pro 推理成本 · 推理成本
so on DS it'd be ≈$0.08
tags: 好数字·好思考 · → daily
value: qty=≈$0.08
- <a id="atom-f95d0fd1466e7f64"></a>🟥 2026-06-01
群 [老旧] · 200M 营收却能训练前沿模型存疑 · 营收/模型训练成本
how does deepseek pay for itself? They have 200m of revenuve and are training models near the frontier (with NVDA chips)
tags: 好问题·好数字 · → daily
value: qty=$200M
- <a id="atom-1b052494b8819984"></a>🟥 2026-06-27
X·@Pier4r · API 定价 · 定价权/价格动态
API prices are profitable but 1/2
tags: 好观点·好思考 · → daily
value: qty=亏损 · currency=USD · date=2026
- <a id="atom-d3b2f1089d65f573"></a>🟥 2026-06-26
X·@bdsqlsz · Wenfeng misjudged agent trend: chose general model only · 技术路线/竞争格局
因为文锋误判了agent的风潮。24年他认为中国的saas市场不够大,写代码的程序员没有那么多,所以只选择做通用模型。
tags: 好观点·好思考 · → daily
- <a id="atom-32528e93d5be5994"></a>🟥 2026-05-10
X·@teortaxesTex [老化中] · 训练数据可被通过提问窃取 · 数据安全/模型安全
You Can Steal DeepSeek Training Data By Questioning This
tags: 好信源·好问题 · → daily
📷 原图
- <a id="atom-d6ede899ed73c46a"></a>🟥 2026-04-28
X·@teortaxesTex [老化中] · 员工薪酬状况 · 融资/人事变动
Wenfeng has got to raise money asap, employees can't even afford premium… brain drain accelerating…
tags: 好观点·好思考 · → daily
value: qty=cannot afford premium · direction=low
- <a id="atom-316dd9f84c87747f"></a>🟥 2026-04-28
X·@teortaxesTex · 融资需求 · 融资
Wenfeng has got to raise money asap
tags: 好观点·好思考 · → daily
value: qty=asap · date=2026-04-28
- <a id="atom-680fe3a68879d631"></a>🟥 2026-04-17
X·@teortaxesTex · 收入潜力低于3亿美元 · 营收预测/开源策略
I don't see how he can get even $300M, except as charity.
tags: 好观点·好思考 · → daily
value: qty=小于3亿美元 · direction=less_than
- <a id="atom-c04380a71f799b26"></a>🟥 2026-06-27
X·@teortaxesTex · 推理技术更新频率 · 技术路线
It's almost 2 months old, they have probably moved on to the next big thing
tags: 好思考 · → daily
value: date=2026-06-27T12:37:15
📷 原图
- <a id="atom-4fe49c51402e0341"></a>🟥 2026-07-01
X·@ChrisRMcGuire · owns so few chips · 供给产能/竞争格局
tags: 好数字 · → daily
- <a id="atom-daa02881aa629988"></a>🟥 2026-04-30
X·@Presidentlin · impact on Anthropic timeline · 竞争格局
DeepSeek moved us probably later this month, which will bring it to 3 months
tags: 好数字 · → daily
- <a id="atom-d610a7b8102c1dd2"></a>🟥 2026-04-23
X·@ZhihuFrontier · 推理速度 · 技术路线
Only around 20 TPS, noticeably slow
tags: 好数字 · → daily
value: qty=~20 TPS
📷 原图
- <a id="atom-aef80f427fbcdbaf"></a>🟥 2026-04-20
X·@ChinaMacroFacts · 年烧钱规模 · 资本开支
50亿26年不够烧一年
tags: 好数字 · → daily
value: qty=$5B · date=2026
- <a id="atom-972e88499dc90207"></a>🟥 2026-06-27
X·@scaling01 · 推理效率排名(低于GLM) · 推理效率
_so reasoning efficiency:
Kimi > GLM > DeepSeek_
tags: 好观点 · → daily
- <a id="atom-d1b7258d18b4dd4a"></a>🟥 2026-05-25
X·@agupta · 前沿模型定价 · 价格动态/竞争格局
deepseek just permanently priced their frontier model at 1/30th of american labs
tags: 好信源 · → daily
value: qty=1/30th of American labs
- <a id="atom-01e7ce886f1a0e2f"></a>🟥 2026-05-25
X·@theinformation · 商业化转型驱动因素 · 资本开支
rising computing costs push the once research-focused lab toward commercialization
tags: 好思考 · → daily
value: direction=rising
- <a id="atom-5285cf6ac6c9749a"></a>🟥 2026-05-17
X·@Linahuaa · 自动拒绝美国本科中国申请者 · 人事变动
DeepSeek is reportedly auto-rejecting Chinese applicants who went to America for undergraduate.
tags: 好信源 · → daily
📷 原图
- <a id="atom-036372a2912307ac"></a>🟥 2026-04-26
X·@teortaxesTex [老化中] · ex-employees from frontier AI labs · 竞争格局
no frontier lab alumni
tags: 好信源 · → daily
value: qty=no · date=2026-04-26
- <a id="atom-c3f81ab6cc9bcfed"></a>🟥 2026-04-26
X·@Presidentlin [老化中] · 当前竞争优势 · 竞争格局
You guys are really kicking DeepSeek butts for now.
tags: 好信源 · → daily
value: direction=negative
📷 原图
- <a id="atom-0e5bc27b6882a412"></a>🟥 2026-06-23
X·@georgejrjrjr · 可能失去了优秀人才且未充分补替 · 竞争格局
either the talent they lost may not have been adequately replaced
tags: 好思考 · → daily
- <a id="atom-3795c20d06d6561a"></a>🟥 2026-06-23
X·@georgejrjrjr · 可能被架构问题困扰而竞争者在数据处理上领先 · 技术路线
they stayed nerd sniped by architecture when the competition pulled ahead by focusing on data
tags: 好思考 · → daily
- <a id="atom-8aa45c9264b57928"></a>🟥 2026-06-17
X·@yifever · 估值 · 估值
deepseek is valued less than cursor
tags: 好思考 · → daily
value: qty=估值低于 Cursor
- <a id="atom-e18e086c3ac6406b"></a>🟥 2026-06-16
X·@teortaxesTex · 投资者对无投票权股票态度 · 治理结构/估值
no, investors are not fine with no voting rights, which is why DeepSeek is worth only $50B and not four times that.
tags: 好观点 · → daily
📷 原图
- <a id="atom-87c0306d3a3cd131"></a>🟥 2026-06-16
X·@teortaxesTex · 创始人控制权与股东价值损害 · 治理结构/管理层表态
Wenfeng maintains absolute control so that he can keep doing things destructive to shareholder value with no opposition.
tags: 好观点 · → daily
value: direction=negative
- <a id="atom-403ce79e29929f7b"></a>🟥 2026-06-11
X·@baoshu88 · 融资进行中并酝酿降价 · 融资/价格动态
国内Deepseek融资马上close,为了抢市场份额正酝酿进一步降价,加推自己的harness。
tags: 好思考 · → daily
- <a id="atom-bc1ad50345e14fa3"></a>🟥 2026-06-10
X·@teortaxesTex · 开放源码安全风险 · 地缘政治
This is a hopeless security posture when the adversarial nation has superior closed source AGI built to wreck your shit.
tags: 好思考 · → daily
- <a id="atom-278fd833dde071eb"></a>🟥 2026-06-08
X·@pstAsiatech · pricing triggers price war across China's tech sector · 价格动态/竞争格局
DeepSeek’s bargain AI triggers price war across China’s tech sector
tags: 好观点 · → daily
- <a id="atom-7bfca6a933b08886"></a>🟥 2026-06-08
X·@pstAsiatech · pricing shock pressuring rivals and unsettling cloud providers · 价格动态/竞争格局
DeepSeek’s pricing shock is rippling through China’s AI sector, pressuring rivals and unsettling cloud providers
tags: 好观点 · → daily
🟥 中性 takes (neutral) (20)
- <a id="atom-d87cd20e98324883"></a>🟥 2026-06-29
X·@theinformation · CEO观点:为保持长期竞争力需大规模资金储备 · 融资/竞争格局
If DeepSeek were to stay in the game... he really needed to build a massive war chest, at least in the realms of like tens of billions of dollars to begin with.
tags: 好观点·好信源 · → daily
value: qty=tens of billions of dollars · direction=至少
- <a id="atom-f6d426cc1ede367f"></a>🟥 2026-06-10
X·@NikoMcCarty · 相对于美国公司的时间差距 · 竞争格局
They are content for now to remain ~6 months behind U.S. companies while maintaining a lower profile and team size.
tags: 好观点·好数字 · → daily
value: qty=~6 months behind
📷 原图
- <a id="atom-7fd7963fb4880901"></a>🟥 2026-04-25
X·@teortaxesTex · 下一步行动-内蒙古数据中心 · 供给产能/资本开支
complete their Inner Mongolia datacenters, proceed to train a simplified, larger V5 with 100T+ tokens
tags: 好观点·好数字 · → daily
- <a id="atom-1a4389864b9a332d"></a>🟥 2026-04-25
X·@teortaxesTex · IPO稀释率推测 · 融资
Probably not at the speculated dilution (<15%)
tags: 好数字·好观点 · → daily
value: qty=<15%
- <a id="atom-321308755a00d46d"></a>🟥 2026-04-24
X·@hsu_steve · v4最终训练成本估计 · 资本开支/模型训练
DS didn't report cost of v4 final training run, but could be $15-30M
tags: 好数字·好观点 · → daily
value: qty=$15-30M · date=v4
- <a id="atom-2afbfa2040c5641e"></a>🟥 2026-05-22
X·@zephyr_z9 · 融资规模从7B增至10B · 融资
Looks like funding round increased from $7B to $10B
tags: 好数字 · → daily
value: qty=$10B · date=当前
- <a id="atom-4033be48aee52ff0"></a>🟥 2026-05-02
X·@zephyr_z9 [老化中] · 用户每日使用成本对比 · 模型成本/使用体验
Using something like Claude or GPT would cost me over $600-$1000/day
tags: 好数字 · → daily
value: qty=$600-$1000 · date=daily
- <a id="atom-edc3742b202bade7"></a>🟥 2026-04-20
X·@yicaichina [老化中] · reported opening door for outside investors · 融资/人事变动
DeepSeek's Reported Door Opening for Outside Investors May Be to Retain Top Talent, Insider Says
tags: 好观点·好信源 · → daily
- <a id="atom-e632205e21ae13ff"></a>🟥 2026-06-30
X·@teortaxesTex · 大型稀疏MoE比同等活跃参数密集Transformer更难训练 · 技术路线
large sparse MoEs are harder to train than a dense transformer equivalent of their active count
tags: 好观点·好思考 · → daily
📷 原图
- <a id="atom-8d6d9c9342b0d59d"></a>🟥 2026-06-21
X·@teortaxesTex · GRPO label association · 技术路线/定价策略
It's funny how associated DeepSeek is with the GRPO brand
tags: 好思考·好观点 · → daily
📷 原图
- <a id="atom-cd45ff3a87d155b8"></a>🟥 2026-06-21
X·@teortaxesTex · future PPO adoption (if needed) · 技术路线
They will adopt modern PPO formulations if needed
tags: 好思考·好观点 · → daily
- <a id="atom-32be97b8589e8b0d"></a>🟥 2026-06-14
X·@teortaxesTex · Harness 产品定位 · 产品定位
pretty certain that DeepSeek Harness won't be a Claude Code/Codex/OpenClaw clone
tags: 好观点·好思考 · → daily
📷 原图
- <a id="atom-7e21b198f4d8994a"></a>🟥 2026-06-14
X·@teortaxesTex · 招聘方向 · 人事变动
I mean they're looking for out-there projects
tags: 好观点·好思考 · → daily
- <a id="atom-cf3894ef87cc1c28"></a>🟥 2026-06-11
X·@ZhihuFrontier · open-source philosophy compared to Qwen · 技术路线/竞争格局
DeepSeek is almost the opposite [of Qwen]. Its releases are not always loud... but once the paper dropped, the 58-page report explained everything clearly
tags: 好观点·好思考 · → daily
📷 原图
- <a id="atom-e1df030a75d63615"></a>🟥 2026-05-27
X·@teortaxesTex · V4 项目核心挑战 · 技术路线/模型发布
but being smart across a long task by reading a codebase, planning changes, keeping track of what you're doing across thousands of steps that's a different problem. nobody solved it by making things cheaper.
tags: 好观点·好思考 · → daily
📷 原图
- <a id="atom-d52f6a2a79540b7e"></a>🟥 2026-05-14
X·@NathanCalvin · 对高端芯片需求依然存在但部分被非法转移的 NVIDIA 芯片填补 · 出口管制/供给产能_
In the time since these comments, their need remains, but has been filled partially by high performance NVIDIA chips that were unlawfully diverted.
tags: 好观点·好思考 · → daily
value: direction=na
📷 原图
- <a id="atom-50fb209b9ed5c4f4"></a>🟥 2026-05-06
X·@AndrewCurran_ · 融资金额 · 融资
it is expected raise as much as $4 billion
tags: 好数字 · → daily
value: qty=$4 billion · date=first funding round
📷 原图
- <a id="atom-af343798233ee983"></a>🟥 2026-04-30
X·@teortaxesTex [老化中] · 第三方训练对知识能力的影响 · 技术路线
third-party training predictably hurts knowledge
tags: 好观点·好思考 · → daily
📷 原图
- <a id="atom-728fadd3a1a62af3"></a>🟥 2026-04-28
X·@teortaxesTex [老化中] · 模型特性 · 模型能力/竞争格局
DeepSeek makes the most hallucinatory smart models, also horrible at pushing back on the user's bullshit
tags: 好思考·好观点 · → daily
value: direction=最严重
📷 原图
- <a id="atom-f5b66b00530d9c05"></a>🟥 2026-04-27
X·@teortaxesTex · 模型Mimo-V2.5小于256K上下文时表现可接受 · 技术路线
On <256K it's probably ok
tags: 好观点·好思考 · → daily
📷 原图
🟦 新闻流 (squawk · 39)
展开新闻流 (FirstSquawk / financialjuice / wallstengine / DeItaone — 快讯, 非原创 take)
- <a id="atom-fd48b4733d9dffd6"></a>🟦 2026-06-26
X·@FirstSquawk · workforce expansion plan · 人事扩张
DEEPSEEK LOOKS TO DOUBLE WORKFORCE AS FRESH FUNDING FUELS GROWTH-WSJ
tags: 好信源 · → daily
value: qty=double current workforce · date=ongoing
- <a id="atom-ff5071c132c754b4"></a>🟦 2026-06-26
X·@FirstSquawk · 招聘计划 · 人才竞争/技术路线
DEEPSEEK PLANS HIRING SPREE IN ESCALATION OF CHINA’S AI TALENT WAR-FT
tags: 好信源 · → daily
value: qty=招聘热潮 · date=2026-06-26
- <a id="atom-aea10cbf2eb29374"></a>🟦 2026-06-25
X·@financialjuice · 人员计划 · 人事变动
Deepseek plans to double staff across all departments
tags: 好数字 · → daily
value: qty=翻倍 · date=2026-06-25
- <a id="atom-251a801b25c651f8"></a>🟦 2026-06-25
X·@DeItaone · 计划扩招 · 人事变动/供给产能
*DEEPSEEK PLANS TO DOUBLE STAFF ACROSS ALL DEPARTMENTS
tags: 好思考 · → daily
value: qty=double staff · date=未指定
- <a id="atom-0843a3c6e61d092b"></a>🟦 2026-06-25
X·@FirstSquawk ⭐ · 计划全员扩招翻倍 · 人事变动/融资/资本战略
DeepSeek Planning To Double Staff Across All Departments
tags: 好观点·好信源·好思考 · → daily
value: qty=double · date=2026-06-25
- <a id="atom-8302c875c8e77cf0"></a>🟦 2026-06-25
X·@wallstengine · 计划扩招 · 人事变动
DEEPSEEK PLANS TO DOUBLE STAFF ACROSS ALL DEPARTMENTS
tags: 好数字 · → daily
value: qty=2倍 · date=未指定
- <a id="atom-43ce18ed70b832fb"></a>🟦 2026-06-25
X·@FirstSquawk · headcount plan · 人事变动
DeepSeek Looks to Double Workforce as Fresh Funding Fuels Growth-WSJ
tags: 好信源 · → daily
value: qty=double workforce · date=2026
- <a id="atom-ed18385ab1a48944"></a>🟦 2026-06-20
X·@FirstSquawk · 被美国列入黑名单 · 出口管制/地缘政治
US HOLDS OFF BLACKLISTING DEEPSEEK, 100+ CHINESE FIRMS DESPITE NATIONAL SECURITY CONCERNS: REUTERS
tags: 好信源 · → daily
- <a id="atom-e739d5c371b21183"></a>🟦 2026-06-18
X·@FirstSquawk · tells investors no poaching of people · 人事变动/人才竞争
NO POACHING’ OUR PEOPLE, CHINA'S AI BEHEMOTH DEEPSEEK REPORTEDLY TELLS INVESTORS
tags: 好信源 · → daily
- <a id="atom-459f10a285ead259"></a>🟦 2026-06-17
X·@wallstengine · 黑名单状态 · 出口管制/监管政策
U.S. Holds Off Blacklisting China's Deepseek, Cxmt, Over 100 Other Companies Deemed National Security Risks: Sources - Reuters
tags: 好信源 · → daily
- <a id="atom-a8c20269fa8c124f"></a>🟦 2026-06-16
X·@FirstSquawk · 融资规模 · 融资
DEEPSEEK CLOSES $7B-PLUS FUNDING: INFORMATION
tags: 好数字 · → daily
value: qty=$7B+ · date=2026-06-16
- <a id="atom-97f896d7adf21cbf"></a>🟦 2026-06-16
X·@FirstSquawk · 估值 · 融资
DEEPSEEK FUNDING VALUES CO AT OVER $50B: INFORMATION
tags: 好数字 · → daily
value: qty=$50B+ · date=2026-06-16
- <a id="atom-ae5a821be477c927"></a>🟦 2026-06-16
X·@financialjuice ⭐ · 融资金额 · 融资
DeepSeek secures record $7 billion-plus funding with unconventional deal structure: The Information
tags: 好数字·好信源 · → daily
value: qty=$7 billion-plus · date=2026-06-16
- <a id="atom-1017742d29299679"></a>🟦 2026-06-16
X·@FirstSquawk · 融资规模 · 融资
DeepSeek raises more than $7B in record round featuring unconventional structure that keeps CEO in control.
tags: 好数字 · → daily
value: qty=$7B · date=2026-06-16
- <a id="atom-f299505f6ffe8475"></a>🟦 2026-06-16
X·@wallstengine · 融资金额 · 融资
DEEPSEEK $7.4B FUNDING DETAILS, PER THE INFORMATION: Raised: $7.4B+
tags: 好数字 · → daily
value: qty=$7.4B+
📷 原图
- <a id="atom-ec62e9a1b216586a"></a>🟦 2026-06-16
X·@wallstengine · 估值 · 融资
Valuation: $50B+
tags: 好数字 · → daily
value: qty=$50B+
📷 原图
- <a id="atom-0b711489899d362c"></a>🟦 2026-06-16
X·@wallstengine · 投资者锁定期 · 融资
Investor lockup: 5 years
tags: 好数字 · → daily
value: qty=5 years
📷 原图
- <a id="atom-02f476bc52d0236a"></a>🟦 2026-06-16
X·@wallstengine · 投票权 · 融资
Voting rights: mostly none
tags: 好数字 · → daily
value: qty=none
📷 原图
- <a id="atom-b57733e74e4f4d2f"></a>🟦 2026-06-16
X·@wallstengine · 中国国家人工智能基金直接投资 · 融资
Exception: China’s national AI fund invests directly
tags: 好信源 · → daily
📷 原图
- <a id="atom-057537550f79f5ce"></a>🟦 2026-06-16
X·@wallstengine · 战略投资者 · 融资
Strategic backers reportedly include Tencent, CATL, JD, NetEase and IDG.
tags: 好信源 · → daily
value: qty=['Tencent', 'CATL', 'JD', 'NetEase', 'IDG']
📷 原图
- <a id="atom-9ca57b29d51abf3d"></a>🟦 2026-06-16
X·@financialjuice · 被美国列入黑名单延迟 · 出口管制
U.S. delays blacklisting China's DeepSeek
tags: 好信源 · → daily
value: date=近期
- <a id="atom-b1dd31cdeefdd2cb"></a>🟦 2026-06-05
X·@wallstengine ⭐ · U.S. business spending index rank in June · 订单需求/市场份额
DeepSeek took the top spot on a major U.S. business spending index in June
tags: 好数字·好信源 · → daily
value: date=June · direction=top
- <a id="atom-758113b16d578d8b"></a>🟦 2026-06-03
X·@wallstengine ⭐ · 首次融资额 · 融资
DeepSeek is set to raise about 50B yuan, or $7.4B, in its first funding round, per Reuters.
tags: 好数字·好信源 · → daily
value: qty=$7.4B · date=2026-06-03
📷 原图
- <a id="atom-4b980b6ba4893be7"></a>🟥 2026-06-03
X·@wallstengine ⭐ · 投后估值区间 · 估值
The round could value the Chinese AI startup at $52B-$59B post-money.
tags: 好数字·好信源 · → daily
value: qty=$52B-$59B · date=2026-06-03
📷 原图
- <a id="atom-3984febadeb848e2"></a>🟦 2026-05-22
X·@FirstSquawk · 融资目标金额 · 融资
DEEPSEEK IS SET TO CLOSE ITS FIRST FUNDING ROUND SOON, TARGETING AROUND 50 BILLION YUAN.
tags: 好数字 · → daily
value: qty=50 billion · currency=CNY · direction=raising
- <a id="atom-e5c3f10e6f78a0d7"></a>🟦 2026-05-08
X·@FirstSquawk · CEO Liang intent to offer financial support · 融资
DEEPSEEK'S CEO LIANG INTENDS TO OFFER FINANCIAL SUPPORT TO THE COMPANY.
tags: 好信源 · → daily
- <a id="atom-928e90f13d4a0091"></a>🟦 2026-05-08
X·@FirstSquawk · V4模型更新计划 · 产品发布/路线图
DEEPSEEK PLANS TO LAUNCH AN UPDATE FOR ITS V4 MODEL IN JUNE.
tags: 好信源 · → daily
value: date=June
- <a id="atom-04e600c820d69115"></a>🟦 2026-05-08
X·@DeItaone · 募资目标 · 融资
DEEPSEEK AIMING TO RAISE UP TO 50B YUAN: THE INFORMATION
tags: 好数字·好信源 · → daily
value: qty=50B · currency=RMB
- <a id="atom-7f8b3f511466f99c"></a>🟦 2026-05-08
X·@DeItaone · V4模型更新计划 · 产品发布
DEEPSEEK AIMS TO RELEASE UPDATE TO V4 MODEL IN JUNE:INFORMATION
tags: 好信源 · → daily
value: date=June 2026
- <a id="atom-e8d0bef87dce4bc5"></a>🟦 2026-05-08
X·@wallstengine · 融资目标金额 · 融资
DeepSeek is aiming to raise up to $7.35B
tags: 好数字 · → daily
value: qty=$7.35B · date=2026
📷 原图
- <a id="atom-2ae154fe00f7d4dd"></a>🟦 2026-05-08
X·@wallstengine · 投后估值 · 融资
The round could value DeepSeek at more than $51.5B
tags: 好数字 · → daily
value: qty=$51.5B · date=2026
📷 原图
- <a id="atom-f3833f90b2c4e0dc"></a>🟦 2026-05-06
X·@financialjuice ⭐ · 估值 · 估值
Deepseek nears $45B valuation - FT
tags: 好数字·好信源 · → daily
value: qty=$45B · date=2026-05-06
- <a id="atom-20f573a3f04936f3"></a>🟦 2026-05-06
X·@FirstSquawk ⭐ · 估值 · 融资
DEEPSEEK NEARS $45BN VALUATION AS CHINA'S 'BIG FUND' LEADS INVESTMENT TALKS
tags: 好数字·好信源 · → daily
value: qty=$45BN · date=2026-05-06
- <a id="atom-98f346a44a76bc96"></a>🟦 2026-05-06
X·@FirstSquawk · 正在进行融资谈判 · 融资
DEEPSEEK NEARS $45BN VALUATION AS CHINA'S 'BIG FUND' LEADS INVESTMENT TALKS
tags: 好信源 · → daily
value: direction=in progress
- <a id="atom-bcc8c74bfce8ebdb"></a>🟦 2026-05-06
X·@FirstSquawk · 估值 · 估值
DeepSeek nears $45B valuation
tags: 好数字 · → daily
value: qty=$45B · date=2026-05-06
- <a id="atom-79c86ee887fe7ad9"></a>🟦 2026-05-06
X·@wallstengine [老化中] · 第一轮融资估值 · 融资
FT: DeepSeek is in talks for its first fundraising at about a $45B valuation
tags: 好信源 · → daily
value: qty=约450亿美元 · date=2026-05
📷 原图
- <a id="atom-9956d0741ce87ae8"></a>🟦 2026-05-06
X·@wallstengine · 近期估值 · 估值
The AI lab was valued around $20B only weeks ago.
tags: 好数字 · → daily
value: qty=约200亿美元
📷 原图
- <a id="atom-1ec202f4cca20e50"></a>🟦 2026-04-27
X·@wallstengine · V4-Pro model output price cut · 价格动态
DeepSeek cut prices on its new V4-Pro model by 75% through May 5
tags: 好数字 · → daily
value: qty=75% · date=2026-05-05
[图: DeepSeek-V4系列模型降价及促销活动的价格对比表 — DeepSeek-V4-Pro Output价格: $0.87; DeepSeek-V4-Pro Input (cache miss)价格: $0.435; DeepSeek-V4-Pro Input (cache hit)价格: $0.003625; DeepSeek-V4-Pro 促销
📷 原图
- <a id="atom-e5b171850dd2efc1"></a>🟦 2026-04-17
X·@wallstengine ⭐ · 融资目标金额 · 融资
DeepSeek is in talks to raise outside capital for the first time, seeking at least $300M at a $10B+ valuation
tags: 好数字·好信源 · → daily
value: qty=$300M · date=2026
📷 原图
⏱ 时间轴 (近 20)
- 🟥 2026-07-09 ·
narrative · 事件类比 · → daily
- 🟥 2026-07-05 ·
narrative · 内存需求趋势 · → daily
- 🟥 2026-07-03 ·
forecast · 模型发布能力预测 · → daily
- 🟥 2026-07-03 ·
forecast · 成就预期解锁时间 · → daily
- 🟥 2026-07-03 ·
narrative · GPU 资源充足程度 · → daily
- 🟥 2026-07-02 ·
narrative · 招聘战略表明从研究实验室转向产品公司 · → daily
- 🟦 2026-07-02 ·
fact · 招聘Agent Harness角色 · → daily
- 🟥 2026-07-02 ·
narrative · 招聘Agent Infra角色指向平台雄心 · → daily
- 🟦 2026-07-02 ·
fact · 之前网站引导用户使用Claude Code访问agent · → daily
- 🟦 2026-07-02 ·
fact · 招聘普通前端、后端和测试工程师 · → daily
- 🟦 2026-07-02 ·
fact · 计划每个部门人数翻倍 · → daily
- 🟥 2026-07-02 ·
narrative · building AGI · → daily
- 🟥 2026-07-02 ·
narrative · 战略转向: 从研究实验室到产品公司 · → daily
- 🟦 2026-07-02 ·
fact · 仍在招聘普通前端、后端和测试工程师 · → daily
- 🟦 2026-07-02 ·
fact · popularised SPARSE attention starting with DeepSeek V3.2 · → daily
- 🟥 2026-07-02 ·
narrative · 研发策略 · → daily
- 🟦 2026-07-02 ·
fact · Web Traffic排名 · → daily
- 🟦 2026-07-02 ·
fact · AI属性 · → daily
- 🟦 2026-07-02 ·
fact · 运营模式 · → daily
- 🟦 2026-07-02 ·
fact · 招聘优先级 · → daily
📊 板块级信号 · AI labs (84)
*以下是 AI labs 板块的信号 (DeepSeek 是成员), 非直接针对 DeepSeek — 供板块上下文参考*
展开板块信号
- 🟥▼ 2026-06-26 [twitter] · IPO timing
all AI labs IPOing at once
- 🟥▲ 2026-04-23 [twitter] · prompting到AGI路径
AI labs are literally prompting their way to AGI
- 🟥▼ 2026-04-15 [twitter] · job loss actions
jobs losses are also predictable now that AI labs have started hill-climbing powerpoint, excel, word and specific profes
- 🟥▼ 2026-05-13 [twitter] · revenue growth will slow hard due to token optimization
when that happens rev growth for labs will slow hard
- 🟥▼ 2026-06-25 [twitter] · model release pace slowing
This does not slow development in any way, it only slows the rate at which the labs can release models, not how fast the
- 🟦▼ 2026-06-25 [twitter] · internal capability gap widening
The gap between what is available to the public, and what the labs have internally, will steadily widen from this day fo
- 🟥▼ 2026-05-27 [twitter] · become optimal destination to solve customer problems
when they become the optimal destination to solve the problems of your customers, it’s going to get ugly
- 🟥▲ 2026-05-07 [twitter] · 模型供给策略
more diversification in the next 6 months or so as various economically useful capabilities democratize effectively to c
*以下是 Chinese AI labs 板块的信号 (DeepSeek 是成员), 非直接针对 DeepSeek — 供板块上下文参考*
展开板块信号
- 🟥▲ 2026-06-13 [twitter] · 价格竞争力
There’s also really competitive offering from chinese labs out, quality is usually a year behind latest gen but for 10%
- 🟦▲ 2026-06-07 [twitter] · gap 10 years ago
Chinese labs were 10 years behind (per The Economist), now they are 6 months
- 🟥▼ 2026-06-10 [twitter] · 面临美国网络攻击风险
They want to rugpull Chinese labs – no, the entire Chinese IT, if not the whole infrastructure of the society – near the
- 🟥▲ 2026-06-07 [twitter] · gap to frontier closed models
the difference is at max 6 months
- 🟥▲ 2026-06-07 [twitter] · fundamental capability gap
There is no fundamental human resource capability gap. It is compute and $$$ and datasets.
- 🟥▼ 2026-06-04 [twitter] · 停止发布开放权重模型的影响
If they stop, the frontier falls further and further behind to those who want to use local/fine-tuned models
- 🟥▲ 2026-05-31 [twitter] · resource advantages relative to US labs
cheaper energy, more abundant AI researchers and the ability to distill and learn from leading US models, but less capab
- 🟥▼ 2026-07-01 [twitter] · would not economically survive Anthropic competition
Chinese AI labs would not economically survive the juggernaut of Anthropic - unless China took drastic steps
*以下是 AI companies 板块的信号 (DeepSeek 是成员), 非直接针对 DeepSeek — 供板块上下文参考*
展开板块信号
- 🟥▼ 2026-06-12 [twitter] · 投资风险
investing in AI companies has just become permanently more risky as the USG could pull the plug at any moment
- 🟥▲ 2026-06-05 [twitter] · 乐于捐赠股份
And founders / companies are all excited to give! They love the idea of direct philanthropy / private ownership of thei
- 🟥▲ 2026-06-11 [twitter] · improved margins despite open source
They’ve been able to improve margins & manage around open source anyways
- 🟥▲ 2026-05-14 [twitter] · focus on pure maths
Talking about the surge of AI research into pure maths, a friend put it a good way: "unfathomable resources from the big
- 🟥▼ 2026-07-03 [twitter] · 正在失去定价权
AI companies are losing pricing power with increasingly cost-sensitive customers
- 🟥▲ 2026-06-05 [twitter] · 预计有大公告
Expect some big announcements!
- 🟥▲ 2026-05-28 [twitter] · 合规成本可能低于从芝加哥获得的收入
the compliance costs are likely to be smaller than the revenue AI companies can make from just Chicago
- 🟥▲ 2026-05-26 [twitter] · 商业模式持续性
What’ll really happen is AI will become normalised, affordable and integrated into society once the market learns which
← 实体目录 · 系统日志