DeepSeek is a Chinese artificial intelligence startup based in Hangzhou, founded at the end of 2023 by Liang Wenfeng, a serial entrepreneur who also runs the High-Flyer hedge fund Al JazeeraTime. The company develops open-source large language models (LLM) and is owned and funded exclusively by the Chinese hedge fund High-Flyer. DeepSeek - Wikipedia, la enciclopedia libre
DeepSeek announced V3.1 on August 19, 2025, an update to its V3 model that introduces significant improvements, particularly a longer context window of 128,000 tokens, which allows it to consider a greater amount of information per query and maintain longer conversations with better memory. The ZviDigit
Main Features of V3.1
Massive Scale The model has 685 billion total parameters, an increase from its predecessor's 671B, using a Mixture-of-Experts (MoE) architecture that activates only 37 billion parameters per token. DigitMarkTechPost
Hybrid Inference Mode V3.1 introduces a system with two routes: deep reasoning and fast response, which can be chosen through a switch present in both the web version and the app. Genbeta | Geeky Gadgets
Enhanced Capabilities
- Obtained 71.6% on the Aider coding test, placing it above Claude Opus 4. Deepseek lanza su nuevo modelo 3.1
- Greater thinking efficiency: DeepSeek-V3.1-Think achieves response quality comparable to DeepSeek-R1-0528, while responding more quickly. Hugging Face
- Better tool usage: post-training optimization that improved performance in agent tasks. Hugging Face
Technical Optimization The model is trained using the UE8M0 FP8 scale data format, optimized for next-generation chips. Lo que DeepSeek acaba de lanzar no gustará ni a NVIDIA
Why is it So Novel?
1. Economic Disruption of the AI Market
DeepSeek said it had spent only $5.6 million to power its base AI model, compared to the hundreds of millions or billions of dollars that US companies invest. CNN Business
The impact was immediate: Nvidia lost almost $600 billion in market capitalization following a 17% drop in its shares. Al Jazeera
2. Challenge to US Technological Hegemony
The model surprised the industry by surpassing OpenAI, displacing ChatGPT from the top of the iOS App Store. What Is DeepSeek, the New Chinese OpenAI Rival? | TIME
Additionally, the United States has been restricting AI chips to China for years for national security reasons, making it more notable that DeepSeek achieved its model with less powerful chips. What is DeepSeek, the Chinese AI startup that shook the tech world? | CNN Business
3. Open Source vs. Proprietary Model
DeepSeek released V3.1 under MIT License, allowing anyone to use, modify and share it. What is DeepSeek-V3.1 and Why is Everyone Talking About It? - MarkTechPost
This contrasts with closed models from OpenAI and Google.
4. Competitive Performance
DeepSeek competes directly with the best:
DeepSeek-R1 comparable to OpenAI-o1 in mathematics, code and reasoning. Hugging Face
On MATH-500 it achieved 97.3%, surpassing OpenAI o1-1217 with 96.4%. DataCamp
API Prices:
- DeepSeek R1: $0.55 input / $2.19 output (1M tokens).
- OpenAI o1: $15 input / $60 output. → DeepSeek is 96.4% cheaper. Comparación DeepSeek R1 vs OpenAI o1
Strategic Implications
The Sputnik Moment of AI
Marc Andreessen called DeepSeek R1 the "Sputnik moment of AI", alluding to the beginning of the space race. PBS News
Industry Response
US President Donald Trump said DeepSeek should serve as a "wake-up call" for the American industry to compete with greater focus. Al Jazeera
Limitations and Challenges
- Occasional insertions of Chinese text in code outputs, requiring manual corrections. Gizmochina
- Slower speeds in cloud environments, limiting practicality for real-time. Gizmochina
- Political restrictions: censorship regarding criticism of the Chinese Communist Party. TIME
Conclusion
DeepSeek V3.1 is a seismic shift in global AI. More than a technical improvement, it is a demonstration that:
- Innovation can emerge with limited resources: Achieving results comparable to OpenAI with a fraction of the cost challenges assumptions about the need for massive AI investments.
- Open source can compete with proprietary models: DeepSeek's open nature democratizes access to advanced AI.
- Restrictions can drive innovation: Chip limitations imposed on China have paradoxically led to more efficient solutions.
- AI leadership is contested: American technological supremacy in AI is no longer unquestionable.
This development is not only technically impressive, but has profound geopolitical and economic implications, redefining power dynamics in the global race in personal-use AI and enterprise AI applications.
