DeepSeek reveals the secret behind the V3/R1 inference system for the first time, with a theoretical cost profit margin of up to 545%.
DeepSeek discloses the core technical details of its V3/R1 inference system for the first time and discloses amazing profitability! The official Zhihu account of DeepSeek published an article "Overview of DeepSeek-V3/R1 Inference System", which detailed the key technical information of the system and disclosed its theoretical cost and profit margin data for the first time. The article points out that assuming the cost of GPU rental is $2 per hour, the total daily cost is about $87,072. However, based on the pricing strategy of DeepSeekR1, the theoretical total daily revenue can reach US$562,027, which means its theoretical cost profit margin is as high as 545%! This data fully demonstrates the strong profitability and market competitiveness of the DeepSeekV3/R1 inference system, and also
Mar 12, 2025 pm 02:15 PM
Is DeepSeek R1 also brain overload? Performance declines after overthinking, and less thinking can reduce computing costs by 43%.
Large language models (LLMs) may also face the dilemma of "overthinking" when performing tasks, resulting in inefficiency or even failure. Recently, researchers from institutions such as UC Berkeley, UIUC, ETHZurich, and CMU have conducted in-depth research on this phenomenon and published a paper entitled "The Danger of Overthinking: Examining Reasoning-Action Dilemma in Agent Tasks" (Paper Link: https://arxiv.org/pdf/2502.08235). Researchers found that in real-time interactive environments, LLMs often hesitate between "direct action" and "careful planning." This "overthinking" will cause the model to spend a lot of time building complex action plans, but it is difficult to implement effectively
Mar 12, 2025 pm 02:06 PM
The difference between bean bread and deepseek
The main difference between Doubao and DeepSeek is: 1. Doubao is a keyword search engine that relies on keyword matching; DeepSeek is a semantic search engine that understands the semantics of search requests. 2. The source of bean bag data is wide but the quality is uneven. The DeepSeek data is more authoritative but the coverage may be narrow. 3. The bean bun results are presented in a list form, and DeepSeek may provide richer answers and context information. 4. Bean buns are easy to use, and DeepSeek may require more complex query statements. Therefore, bean bread is suitable for quickly finding information, while DeepSeek is suitable for in-depth semantic search and information mining, and the choice depends on the specific needs.
Mar 12, 2025 pm 01:57 PM
What is deepseek?
The meaning of DeepSeek depends on the context and cannot give a single answer. It may refer to: 1. A software or tool that uses deep learning technology to search or analyze data, and the specific function depends on its design; 2. A research project or organization focused on the fields of deep learning, artificial intelligence or data mining; 3. Virtual characters or characters in novels, games or movies. More context information is required to accurately interpret DeepSeek.
Mar 12, 2025 pm 01:54 PM
How strong is deepseek's thinking ability
DeepSeek has limited but impressive thinking ability, and its "thinking" is based on pattern recognition rather than real understanding. 1. It can perform question-and-answer, text generation, translation and summary, but accuracy depends on training data; 2. It lacks common sense and understanding and is susceptible to ambiguity problems; 3. It cannot conduct logical reasoning and critical thinking, and may produce "illusions"; 4. Its ability is advanced pattern matching and text generation, rather than real human thinking. What makes DeepSeek powerful is its efficient text processing capabilities, not its true cognitive capabilities.
Mar 12, 2025 pm 01:51 PM
ZTE Nubia's embedded DeepSeek version updates, the industry realizes intelligent network search for the first time
ZTE Nubia Nebula AI ushers in a major upgrade! The DeepSeek version is updated to bring a smarter and more convenient search experience. The new version of DeepSeek is the first to realize intelligent network search, automatically determine whether it is connected to the Internet based on the usage scenario, without manual switching, and the search is smoother. The voice interaction function has also been launched, and users can easily start in-depth reasoning and analysis with just voice. The generated content supports historical record viewing and can be saved to Notepad with one click, which is convenient for users to organize and review. This update not only improves the performance and functions of DeepSeek, but also enhances its integration with Nebula AI, creating a more integrated and smarter AI experience for users. At present, the new version of DeepSeek has been launched on the Nubia Z70 Ultra model, and will
Mar 12, 2025 pm 01:45 PM
545%! DeepSeek first disclosed cost profit margin Expert: If it is already a company worth more than 10 billion US dollars in the United States
DeepSeek, a Chinese AI startup, has been "open source" recently. Last Saturday (1st), there was a bigger surprise. It fully revealed the DeepSeek-V3/R1 inference system, which not only disclosed the core optimization solution of its inference system, but also disclosed key data such as cost-benefit rates for the first time, causing industry shocks. DeepSeek released its first article on Zhihu platform last Saturday, announcing the details of model inference cost-profits and disclosed key information on cost-profit rates. If all tokens are calculated based on the pricing of DeepSeek-R1, the theoretical total revenue per day is US$560,000, and the cost-profit rate is 545%. This figure refreshes the profit ceiling in the global AI big model field. According to the official disclosure of DeepSeek,
Mar 12, 2025 pm 01:30 PM
What is the difference between bean bread and deepseek
The core difference between bean bun and DeepSeek is retrieval accuracy and complexity. 1. Doubao is based on keyword matching, simple and direct, with low cost, but low accuracy, and is only suitable for structured data; 2. DeepSeek is based on deep learning, can understand semantics, has high accuracy, but high cost, and is suitable for unstructured data. The final choice depends on the application scenario and resource limitations. If the accuracy requirements are not high, choose bean bags, and if you pursue high precision, choose DeepSeek.
Mar 12, 2025 pm 01:24 PM
The hidden competition of the biological version of DeepSeek, the Chinese model is regarded as a stronger opponent, and Sanofi is betting on $1 billion!
China's AI power leads the world: Baitu Biotech xTrimoV3 biological model surpasses Evo2 The Evo2 biological AI model jointly released by Stanford University and Nvidia has caused heated discussion. However, a detail in its paper highlights the rise of China's AI: Baitu Biotech's xTrimo series of models has been listed as a competitor with larger parameters but not yet open source. Baitu Biotech, a company that is regarded as a pioneer in the basic model of life sciences by the US market, has continued to deepen its efforts in this field since 2020. In October 2024, the xTrimoV3 full-modal biological model launched, which set a new record for the world's largest basic life science AI model with 210 billion parameters. The breakthrough of xTrimoV3 lies in its first
Mar 12, 2025 pm 01:18 PM
Visual enhancement fine-tuning! DeepSeek R1 technology has been successfully migrated to multimodal field and is fully open to source
Big recommendation: Visual-RFT - a visual enhancement and fine-tuning open source project to empower visual language models! The AIxiv column continues to focus on top AI research in the world and has published more than 2,000 academic and technical articles. Welcome to contribute to share your outstanding achievements! Submission email: liyazhou@jiqizhixin.com; zhaoyunfeng@jiqizhixin.comVisual-RFT (VisualReinforcementFine-Tuning) project successfully applied the reinforcement learning and reinforcement fine-tuning (RFT) paradigm based on rule rewards to visual language big model (LVLM), breaking through the previous methods that were limited to text, mathematics and other fields.
Mar 12, 2025 pm 01:12 PM
The simplest DeepSeek local deployment strategy for the entire network: Create an exclusive AI assistant
This article provides a super simple tutorial for DeepSeek local deployment to help you easily create an exclusive AI assistant. No need to rely on cloud services, installation and configuration can be completed on Windows, macOS and Linux systems in just a few steps. DeepSeek has certain hardware requirements (16GB memory and solid state drive are recommended). The installation process is simple and intuitive, and you can easily get started even without technical background. Tutorials cover preparation, installation, configuration, operation and optional continuous learning steps, allowing you to quickly experience the powerful features of DeepSeek's text generation, code writing, and translation. Start your hands now and have your exclusive AI!
Mar 12, 2025 pm 01:09 PM
Alibaba releases and opens the Tongyi Qianwen QwQ-32B inference model, with performance comparable to DeepSeek-R1
Alibaba Cloud’s new open source inference model Tongyi Qianwen QwQ-32B, with performance comparable to DeepSeek-R1! On March 6, Alibaba Cloud officially announced the open source of its latest research and development inference model- Tongyi Qianwen QwQ-32B. The model has achieved significant improvements in math, code and general capabilities, with performance levels comparable to the industry-leading DeepSeek-R1 model. This news stimulated Alibaba's stock price to rise sharply, once rising by more than 8%. Alibaba Cloud confirmed in several authoritative benchmark tests that Qianwen QwQ-32B's performance far exceeds OpenAI-o1-mini and is on par with DeepSeek-R1 in overall performance. What's more remarkable is that the model has greatly reduced its performance while maintaining high performance.
Mar 12, 2025 pm 01:00 PM
Make your mark! DeepSeek jumps to second place in the global AI application rankings
In the latest list of the top 100 global generative AI applications released by Andreessen Horowitz (a16z) in 2025, China's AI technology strength performed well. Among them, DeepSeek's web version application ranks second, second only to ChatGPT, and its rapid rise has attracted much attention. The list covers 50 web applications and 50 mobile applications. DeepSeek quickly ranked among the top web applications in less than two months, which is impressive. ByteDance's Doubao is also ranked tenth among web applications. In addition to DeepSeek and Doubao, Chinese AI applications such as Dark Side of the Moon (11th), Conch Video (12th), and Kuaishou Keling (20th), have also surpassed Sora and Midjou.
Mar 12, 2025 pm 12:45 PM
US media: Chinese tech giants that embrace DeepSeek are even stronger
DeepSeek emerged, stirring up the landscape of Chinese technology giants! This article explains how DeepSeek forces giants such as ByteDance, Alibaba, Tencent and Baidu to re-examine their AI strategies and quickly adjust their product strategies. The giants’ response strategy: The four giants have included the DeepSeek model in product plans or actively evaluated. Tencent and Baidu regard it as their core selling point, which is the first case in the domestic AI field, similar to Meta or Google promoting its AI chatbots with the help of third-party models. Tencent’s “Yuanbao” application was even renamed “Yuanbao (DeepSeek Version)” to highlight this cooperation. Baidu Search and Maps have also added DeepSeek search options to provide users with more choices. ByteDance:
Mar 12, 2025 pm 12:42 PM
Hot tools Tags
Undresser.AI Undress
AI-powered app for creating realistic nude photos
AI Clothes Remover
Online AI tool for removing clothes from photos.
Undress AI Tool
Undress images for free
Clothoff.io
AI clothes remover
Video Face Swap
Swap faces in any video effortlessly with our completely free AI face swap tool!
Hot Article
Hot Tools
vc9-vc14 (32+64 bit) runtime library collection (link below)
Download the collection of runtime libraries required for phpStudy installation
VC9 32-bit
VC9 32-bit phpstudy integrated installation environment runtime library
PHP programmer toolbox full version
Programmer Toolbox v1.0 PHP Integrated Environment
VC11 32-bit
VC11 32-bit phpstudy integrated installation environment runtime library
SublimeText3 Chinese version
Chinese version, very easy to use
Hot Topics
1664
14
1423
52
1321
25
1269
29
1249
24