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
StepFun
35 atoms · 跨 19 天 · 首见 2026-04-26 · 最近 2026-06-30
三色: 🟦 fact 26 · 🟥 take 9 · stance ▲10/▼0/◆14
叙事状态: 🍂 fading (退潮) · 💤 沉寂 · 策展近14d 0 atom · 🐦 X 近14d 2
状态只算低频策展源 (群/卖方); X firehose 仅作背景音量
来源: X 35
时态: fresh:32 · aging:3
标签: 好信源:15 · 好数字:12 · 好观点:10 · 好思考:1
🟨 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
StepFun 自2023年成立起便专注基础与多模态模型
¹¹,近期发布198B多模态模型Step 3.7 Flash,推理速度达400 tok/s
¹¹。该模型在ClawEval-1.1和SimpleVQA Search上排名第一,并在Apache 2.0下开源权重
¹¹。产品受到好评
¹,且能以Anthropic模型十分之一的成本提供类似能力
¹。公司正推进香港IPO,融资近25亿美元,有望成为近年来香港最大科技上市之一
¹¹¹。供应链投资者将其视作终端访问机会而非模型公司
¹,香港上市已是时间问题
¹。分析师预期其将留下更大印记
¹。
🧭 拥挤度 (一人一票): ▲ 6 位作者 (KOL6) vs ▼ 0 位作者
⚖️ 多头 6/6 来自KOL
📄 广泛报道的事实 · 2 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)
- 计划香港IPO — 2 源报道 (含1快讯) ·
X
- 融资规模 — 2 源报道 (含2快讯) ·
X
💢 核心分歧 (2 轴)
1. 开源模型能力突破与西方市场认可错位
*topic: 开源模型突破/西方市场认知 · 1 bull vs 1 bear*
🟢 bullish 侧:
🔴 bearish 侧:
2. 推理速度优势能否弥补模型规模劣势
*topic: 推理速度/模型容量瓶颈 · 1 bull vs 1 bear*
🟢 bullish 侧:
🔴 bearish 侧:
🟦 客观事实 (facts) (26)
- <a id="atom-f92935bd613dac32"></a>🟦 2026-06-30
X·@StepFun_ai ⭐ · 公司成立时间 · 公司成立
We started in 2023, and yes - foundation and multimodal models have been our focus from day one.
tags: 好数字·好信源 · → daily
value: date=2023
- <a id="atom-b0ab74e17c7f953b"></a>🟦 2026-05-29
X·@ivanfioravanti ⭐ · Step 3.7 Flash 发布 · 模型发布
Step 3.7 Flash is here: The new frontier is agent efficiency.
tags: 好数字·好信源 · → daily
📷 原图
- <a id="atom-55c7a354ec363195"></a>🟦 2026-05-31
X·@ZhihuFrontier · 发布Step3.7-Flash · 模型发布
_@StepFun_ai releases Step3.7-Flash, a 198B multimodal model with configurable reasoning levels._
tags: 好信源·好数字 · → daily
- <a id="atom-0462aecad72d46c3"></a>🟦 2026-05-29
X·@kilocode · Step 3.7 Flash 推理速度 · 模型发布/技术路线
running at 400 tok/s
tags: 好数字 · → daily
value: qty=400 tok/s · date=2026-05-29
- <a id="atom-e5efad66fc8b7d16"></a>🟦 2026-05-29
X·@kilocode · Step 3.7 Flash 基准测试排名 · 模型发布/技术路线
#1 ClawEval-1.1 (67.1), #1 SimpleVQA Search (79.2), #2 SWE-PRO (56.3), 95.3 on V* Python. Open weights under Apache 2.0.
tags: 好数字 · → daily
value: qty=#1 on ClawEval-1.1 (67.1), #1 on SimpleVQA Search (79.2), #2 on SWE-PRO (56.3), 95.3 on V* Python
- <a id="atom-0a5bb51d17adf483"></a>🟦 2026-05-08
X·@yicaichina · 融资规模 · 融资
StepFun to Raise Nearly USD2.5 Billion as Chinese AI Startup Advances Hong Kong IPO
tags: 好数字 · → daily
value: qty=~USD2.5B
- <a id="atom-29050ba58c6524aa"></a>🟦 2026-06-30
X·@StepFun_ai · 主要业务方向 · 业务方向
foundation and multimodal models have been our focus from day one.
tags: 好信源 · → daily
- <a id="atom-6a53f588abad5578"></a>🟦 2026-05-29
X·@kilocode · Step 3.7 Flash 模型权重开源 · 模型发布/技术路线
shipped Step 3.7 Flash with the weights and quants ready to go
tags: 好信源 · → daily
value: date=2026-05-29
- <a id="atom-763a12993eb5d42e"></a>🟦 2026-05-29
X·@StepFun_ai · Step 3.7 Flash 在 Kilo 平台可用 · 合作客户/产品发布
love seeing 3.7 flash live in kilo🤝— and indeed 400 tok/s is no joke😎
tags: 好信源 · → daily
value: date=2026-05-29
- <a id="atom-b9b98ac99b49807b"></a>🟦 2026-05-26
X·@pstAsiatech · 私人融资轮 · 融资
AI model startups Moonshot AI and StepFun went ahead with large private funding rounds.
tags: 好信源 · → daily
value: date=May 2026
- <a id="atom-c90d05bb76eeed4a"></a>🟦 2026-06-08
X·@pstAsiatech · 计划香港IPO · IPO
Chinese AI Start-Up StepFun Set to File for Hong Kong IPO
tags: 好数字 · → daily
- <a id="atom-e9625f44c17a8810"></a>🟦 2026-06-03
X·@StepFun_ai · 模型发布 · 模型发布
Deploy Step 3.7 Flash on @modal with SGLang 🚀
tags: 好数字 · → daily
value: qty=Step 3.7 Flash
- <a id="atom-bc191b468bbde7b8"></a>🟦 2026-05-10
X·@poezhao0605 · 融资规模 · 融资
StepFun is reportedly raising $2.5 billion ahead of a possible Hong Kong IPO
tags: 好数字 · → daily
value: qty=$2.5 billion · date=2026-05-10
- <a id="atom-e7e2490ee51377dc"></a>🟦 2026-05-10
X·@poezhao0605 · 传闻融资额 · 融资
StepFun is reportedly raising $2.5 billion
tags: 好数字 · → daily
value: qty=$2.5 billion
- <a id="atom-2f931f238e2b880e"></a>🟦 2026-05-08
X·@poezhao0605 · 融资规模 · 融资
Today's $2.5B round validates the bet.
tags: 好数字 · → daily
value: qty=$2.5B · date=2026-05-08
- <a id="atom-2966fa54708555ea"></a>🟦 2026-05-08
X·@poezhao0605 · 融资规模 · 融资
StepFun (阶跃星辰), a Shanghai-based foundation model company founded by former Microsoft Global VP Jiang Daxin, is reportedly closing a ~$2.5 billion round.
tags: 好数字 · → daily
value: qty=~$2.5B
📷 原图
- <a id="atom-802308fbc2cfbd31"></a>🟦 2026-06-24
X·@CyouSakura · 招聘 researcher 从事 LLM post-training · 人事变动
_Personal Update: I've joined @StepFun_ai as a Researcher, working on LLM post-training._
tags: 好信源 · → daily
- <a id="atom-79b15c4e73e14698"></a>🟦 2026-06-18
X·@StepFun_ai · 与 Cline 合作 · 合作客户
Excited to partner with @cline to bring Step 3.7 Flash to more coding agent builders.
tags: 好信源 · → daily
- <a id="atom-2be3c0e238762a23"></a>🟦 2026-06-16
X·@StepFun_ai · Step 3.7 Flash 发布 · 模型发布
_Excited to see Step 3.7 Flash live via @novita_labs on @OpenRouter._
tags: 好信源 · → daily
- <a id="atom-f2904fdc01a65980"></a>🟦 2026-06-10
X·@DIGITIMESAsia · 准备提交香港IPO申请 · 融资/IPO
Chinese #AI startup StepFun is preparing to file for a Hong Kong #IPO
tags: 好信源 · → daily
- <a id="atom-bf520a94340865a7"></a>🟦 2026-06-04
X·@StepFun_ai · Step 3.7 Flash 模型发布 · 模型发布
_Great to see Step 3.7 Flash live on @FireworksAI_HQ._
tags: 好信源 · → daily
- <a id="atom-c9bb12209aff30f9"></a>🟦 2026-05-28
X·@Xianbao_QIAN · 模型发布 · 模型发布
StepFun 3.7 Flash is out.
tags: 好信源 · → daily
value: date=2026-05-29
📷 原图
- <a id="atom-0f48b006ae0defc5"></a>🟦 2026-05-23
X·@StepFun_ai · StepAudio 2.5 Realtime 发布 · 产品发布
StepAudio 2.5 Realtime is live!
tags: 好信源 · → daily
value: date=2026-05-23
📷 原图
- <a id="atom-ca469b06f0a3b68e"></a>🟦 2026-05-08
X·@poezhao0605 · 投资者构成 · 合作客户/供应链
The investor list tells the story. Huaqin, Longcheer, OmniVision, ZTE. These are phone and consumer electronics supply chain companies locking in model partnerships for device-side AI.
tags: 好信源 · → daily
📷 原图
- <a id="atom-1d9c5b00b36b589e"></a>🟦 2026-06-01
X·@FireworksAI_HQ · 模型架构技术 · 技术路线
Multi-Matrix Factorization Attention (MFA) → KV-cache at ~22% of DeepSeek
tags: 好观点 · → daily
- <a id="atom-b97460985ac66cd0"></a>🟦 2026-05-08
X·@poezhao0605 · 商业模式 · 定价权/技术路线
StepFun's revenue model combines device-side licensing with cloud-based consumption billing. A different path from the API pricing war most Chinese model companies are fighting.
tags: 好观点 · → daily
📷 原图
🟥 多头 takes (bullish) (7)
- <a id="atom-8ec86c8c1593bad8"></a>🟥 2026-05-28
X·@Xianbao_QIAN · IPO计划 · IPO
they're marching towards being IPO listed on Hong Kong.
tags: 好思考 · → daily
📷 原图
- <a id="atom-065eac21169c04c5"></a>🟥 2026-06-24
X·@teortaxesTex · 模型能力预期 · 技术路线
I expect they'll leave a bigger mark
tags: 好观点 · → daily
- <a id="atom-3a62e60c7027b58e"></a>🟥 2026-06-10
X·@DIGITIMESAsia · IPO可能成为近年来香港最大科技上市之一 · IPO/融资
in what could become one of the city's largest technology listings in recent years
tags: 好观点 · → daily
- <a id="atom-5c72f6d57701a396"></a>🟥 2026-05-27
X·@IqraSaifiii · 产品评价 · 产品发布
_@StepFun_ai done a great job_
tags: 好观点 · → daily
- <a id="atom-94d7b7b0dc9ca493"></a>🟥 2026-05-08
X·@poezhao0605 [老化中] · 投资方类型 · 合作客户
Supply chain investors are not funding a model company. They are funding a terminal access play.
tags: 好观点 · → daily
- <a id="atom-abd4424eceab1e38"></a>🟥 2026-05-08
X·@poezhao0605 · 香港IPO预期 · 融资
A Hong Kong IPO is now a matter of when, not whether.
tags: 好观点 · → daily
📷 原图
- <a id="atom-9b49e7e92b9f4173"></a>🟥 2026-04-27
X·@Presidentlin [老化中] · 模型成本与Anthropic比较 · 技术路线
they have models like Anthropics models but at 1/10th the cost
tags: 好观点 · → daily
value: qty=1/10th
🟥 中性 takes (neutral) (2)
- <a id="atom-187ee4a23900220f"></a>🟥 2026-04-27
X·@teortaxesTex · will adopt similar training data scaling approach · 训练数据
StepFun will do the same btw
tags: 好观点 · → daily
📷 原图
- <a id="atom-a04bdb8291991049"></a>🟥 2026-04-26
X·@teortaxesTex [老化中] · model tier status relative to Moonshot and DeepSeek · 竞争格局
StepFun is potentially in the same tier
tags: 好观点 · → daily
value: status=potentially same tier
⏱ 时间轴 (近 20)
- 🟦 2026-06-30 ·
fact · 公司成立时间 · → daily
- 🟦 2026-06-30 ·
fact · 主要业务方向 · → daily
- 🟥 2026-06-24 ·
forecast · 模型能力预期 · → daily
- 🟦 2026-06-24 ·
fact · 招聘 researcher 从事 LLM post-training · → daily
- 🟦 2026-06-18 ·
fact · 与 Cline 合作 · → daily
- 🟦 2026-06-16 ·
fact · Step 3.7 Flash 发布 · → daily
- 🟦 2026-06-10 ·
fact · 准备提交香港IPO申请 · → daily
- 🟥 2026-06-10 ·
narrative · IPO可能成为近年来香港最大科技上市之一 · → daily
- 🟦 2026-06-08 ·
fact · 计划香港IPO · → daily
- 🟦 2026-06-04 ·
fact · Step 3.7 Flash 模型发布 · → daily
- 🟦 2026-06-03 ·
fact · 模型发布 · → daily
- 🟦 2026-06-01 ·
fact · 模型架构技术 · → daily
- 🟦 2026-05-31 ·
fact · 发布Step3.7-Flash · → daily
- 🟦 2026-05-29 ·
fact · Step 3.7 Flash 发布 · → daily
- 🟦 2026-05-29 ·
fact · Step 3.7 Flash 推理速度 · → daily
- 🟦 2026-05-29 ·
fact · Step 3.7 Flash 基准测试排名 · → daily
- 🟦 2026-05-29 ·
fact · Step 3.7 Flash 模型权重开源 · → daily
- 🟦 2026-05-29 ·
fact · Step 3.7 Flash 在 Kilo 平台可用 · → daily
- 🟦 2026-05-28 ·
fact · 模型发布 · → daily
- 🟥 2026-05-28 ·
forecast · IPO计划 · → daily
*以下是 Chinese AI labs 板块的信号 (StepFun 是成员), 非直接针对 StepFun — 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
← 实体目录 · 系统日志