DeepSeek
PulseAugur coverage of DeepSeek — every cluster mentioning DeepSeek across labs, papers, and developer communities, ranked by signal.
- parent of DeepSeek-V3 100%
- parent of DeepSeek V4 100%
- founded by Liang Wenfeng 100%
- parent of DeepSeek-R1 100%
- subsidiary of High-Flyer 100%
- parent of DeepSeek V2 100%
- instance of DeepSeek V4-Pro 95%
- developed V4-Pro 95%
- instance of DeepSeek V4-Flash 95%
- founded Liang Wenfeng 90%
- developed V4 90%
- developed by DeepSeek V4-Flash 90%
- 2026-05-25 product_launch DeepSeek announced a 75% permanent price reduction for its V4-Pro model. 来源
- 2026-05-24 product_launch DeepSeek announced a permanent 75% discount on its flagship AI model. 来源
- 2026-05-24 product_launch DeepSeek announced a permanent 75% discount on its flagship AI model. 来源
- 2026-05-24 product_launch DeepSeek is permanently reducing the price of its V4-Pro AI model by 75%. 来源
- 2026-05-24 funding CATL chairman Zeng Yuqun is investing 10.5 billion yuan in AI computing power, targeting DeepSeek.
- 2026-05-24 funding DeepSeek received a 10.5 billion yuan investment from CATL chairman Zeng Yuqun's firm, Longzhong Dui. 来源
- 2026-05-24 funding CATL chairman Zeng Yuqun invested 10.5 billion yuan in AI computing power leader DeepSeek.
- 2026-05-24 funding DeepSeek received a 10.5 billion yuan investment from CATL chairman Zeng Yuqun.
- 2026-05-24 funding CATL chairman Zeng Yuqun is investing 10.5 billion yuan into AI computing power leaders, including DeepSeek.
- 2026-05-23 product_launch Deepseek permanently adopts a 75% discount on its V4-Pro model pricing. 来源
- 2026-05-22 funding Deepseek is reportedly nearing a $10 billion funding round at a $45 billion valuation. 来源
- 2026-05-22 product_launch DeepSeek AI permanently reduces pricing for its V4 Pro model.
- 2026-05-22 funding DeepSeek is reportedly in discussions for a 70 billion yuan funding round with state-backed investors.
- 2026-05-22 funding DeepSeek is in discussions for a 70 billion yuan funding round with state-backed investors, potentially including 10 billion yuan from a Chinese state AI fund. 来源
- 2026-05-22 funding DeepSeek is reportedly seeking up to 70 billion RMB in a funding round, with potential investors including CATL, JD.com, and NetEase. 来源
26 天有情绪数据
-
百度人工智能业务营收首次占比过半
百度2026年第一季度财报显示，人工智能业务营收达到136亿元人民币，占其核心业务收入的52%，首次成为公司主要的收入来源。公司强调其AI云服务增长强劲，猛增79%，GPU云服务增长184%。百度的自动驾驶服务Apollo Go在本季度也实现了显著增长，提供了超过320万次的全无人驾驶出行服务。
-
中国电信巨头推出 AI 令牌定价计划
中国电信、中国移动和中国联通等中国主要电信公司现已推出基于 AI 令牌的定价套餐。这些套餐超越了传统的蜂窝数据，起价为每月 9.9 元人民币，包含 1000 万个令牌。套餐设计用于各种用途，从日常 AI 任务到更复杂的应用，如编码和 AI 代理部署，企业套餐提供显著更高的令牌额度。
-
China to lead global AI talent pool, surpassing US by 2025
A report by The Economist and the MacroPolo think tank indicates that China is set to dominate the field of top AI researchers. Projections suggest that by 2025, China's AI talent pool could be twice the size of the Uni…
-
Chinese markets surge past 3 trillion yuan turnover amid tech news
The Shanghai and Shenzhen stock markets have seen significant trading volume, with turnover exceeding 3 trillion yuan. This surge in activity is accompanied by various other financial and tech-related news. Notably, Goo…
-
Agentic AI expands across industries, with IT pros reporting high adoption
Agentic AI is rapidly expanding its applications across various industries, with IT professionals reporting widespread adoption and identifying key use cases. Companies like Dell are focusing on balancing the safety and…
-
New PRISM framework corrects SFT flaws in multimodal LLM training
New research from institutions including the Hong Kong University of Science and Technology (Guangzhou) reveals a critical flaw in the common post-training paradigm for multimodal large language models (MLLMs). The stan…
-
开源模型落后于前沿闭源模型，基准测试存在争议
多家领先的 AI 实验室发布了新的开源模型，包括 DeepSeek V4、Gemma 4、Kimi K2.6 和 MiMo 2.5。CAISI 的一项评估表明，这些开源模型落后于前沿闭源模型，且差距正在扩大。然而，评估方法和基准测试的局限性也引发了争议，一些人认为标准化测试未能完全捕捉实际能力，尤其是在编码等复杂任务中。
-
DeepSeek-V4-Flash boosts LLM steering vector research
The release of DeepSeek-V4-Flash has revitalized interest in LLM steering vectors, a method for controlling AI behavior. This new model exhibits exceptional responsiveness to steering instructions, potentially revolutio…
-
Open-weight AI models cost developers fraction of traditional inference
A developer detailed their experience using open-weight AI models for a coding project, incurring a cost of only $5 for over 400 million tokens via a subscription service. This contrasts sharply with the estimated $138.…
-
DwarfStar 4 enables rapid, usable local LLM setups
DwarfStar 4 (DS4) is a new local LLM setup designed for usability and rapid deployment. It aims to make running large language models on personal hardware more accessible. The project highlights the increasing feasibili…
-
DeepSeek V4 推出 MegaMoE 优化以实现高效 MoE
DeepSeek 发布了其 V4 模型，该模型通过一个名为 MegaMoE 的新系统进行了显著优化。该系统使用一个 1400 行的融合 CUDA 内核，通过对模型层内的通信和计算进行细粒度流水线处理来提高性能。这种方法解决了 Mixture-of-Experts (MoE) 模型中通常需要大量 all-to-all 通信的挑战。
-
Cursor IDE users ask about DeepSeek model integration
A user on Reddit is inquiring about the integration of DeepSeek's newer, reportedly efficient models within the Cursor IDE. The question specifically asks if these models can be utilized and if they consume less of the …
-
Redis creator releases DwarfStar 4 for fast local AI inference
DwarfStar 4 (DS4), a new local AI inference engine, has gained rapid popularity for its focus on integrating a single, high-performance model. Developed by Salvatore Sanfilippo, creator of Redis, DS4 is specifically opt…
-
US and China AI ecosystems deeply intertwined despite competition
A new paper from Anthropic and research from arXiv explore the complex relationship between US and Chinese AI development, challenging the notion of a simple race. While the US currently leads in frontier AI, the resear…
-
开发人员在 LLM 应用部署中面临隐藏成本
估算由大型语言模型（LLM）驱动的 AI 应用的部署成本至关重要，因为生产费用可能远远超出最初的预测。开发人员常常低估成本，只关注单个 API 调用，而忽略了用户交互、对话历史和复杂代理工作流的累积费用。输入和输出 token 数量、模型选择、重试率以及检索增强生成（RAG）等技术的使用都会显著影响最终账单，因此需要仔细的架构规划来管理费用。
-
新的大型语言模型因过大或过于复杂而无法在家庭实验室运行
作者详细说明了最近发布的三个大型语言模型——DeepSeek V4-Pro、DeepSeek V4-Flash 和 Zyphra ZAYA1-8B——目前为何无法在典型的家庭实验室硬件上运行。DeepSeek V4-Pro 体积过大，为 805 GB，需要数据中心规模。DeepSeek V4-Flash 虽然体积较小，但仍需要大量内存，并且缺乏广泛的软件支持。Zyphra ZAYA1-8B 体积合适，但使用了新颖的架构，尚未开发出相应的推理软件。
-
OpenAI, DeepSeek, Groq show reliability issues in LLM uptime study
A 30-day monitoring project revealed significant reliability differences among major LLM providers. OpenAI experienced frequent and lengthy outages, while DeepSeek had a concerning number of silent failures that went un…
-
使用本地模型和区块链开发的去中心化AI基础设施
作者详细介绍了开发去中心化AI基础设施的历程，从关于搜索功能的争论开始，发展到解决在线注意力经济中感知到的意义丧失问题。他们构建了一个名为‘iSE DS’的本地AI模型，并将其中的10万份副本分发给中国的DeepSeek，通过将流量和存储转移给用户来减少数据中心的能源浪费。该项目还涉及创建一个升级版的Tor网络和一个基于IPFS的注册表，以增强去中心化的信息访问能力，并计划未来与TON集成以实现全球支付。
-
Tencent open-sources Agent Memory to cut token use by 61%
Tencent has open-sourced its Agent Memory technology, designed to significantly reduce token consumption and improve task success rates for AI agents. This new solution compresses long-term and short-term memory by offl…
-
SalesEase携手腾讯，应对SaaS洗牌与Agent浪潮
Salesforce中国总裁邓勇福讨论了公司与腾讯的战略合作，他认为这对于应对竞争激烈的SaaS格局和新兴的Agent趋势至关重要。此次合作超越了传统云服务，专注于技术、产品和生态系统的深度协同。SalesEase已利用腾讯的支持快速推出Agent产品，包括一款与企业微信和WorkBuddy集成的“销售专属龙虾”，旨在区别于专注于价格战的竞争对手。