Currently ranking as the top free app on Apple's App Store, DeepSeek experienced widespread outages this morning, affecting its API and web chat services. According to the company's status page, the API was operating with "degraded performance," while the web chat service faced a "partial outage."
DeepSeek, a Chinese startup and major AI competitor to ChatGPT, has rapidly emerged as a formidable player in the AI landscape. Founded in 2023, DeepSeek, backed by the hedge fund High-Flyer, focuses on developing open-source large language models (LLMs) that rival or surpass existing industry leaders in performance and cost efficiency.
Established by Liang Wenfeng, who co-founded High-Flyer, DeepSeek is headquartered in Hangzhou, Zhejiang. The company's mission centers on advancing artificial general intelligence (AGI) through open-source research and development. This approach aims to democratize AI technology, making it accessible for commercial and academic applications.
DeepSeek has temporarily restricted new user registrations to individuals with China-based phone numbers. This measure was implemented in response to "large-scale malicious attacks" targeting the platform's online services. Existing users remain unaffected and can continue to access their accounts without interruption.
Initially, DeepSeek announced that only users with phone numbers bearing the +86 country code could register for new accounts. Later, the website's message was updated to indicate that "registration may be busy," suggesting a potential easing of the restriction. At that time, some users reported being able to register using email addresses — I registered this morning using my personal email address.
Open-Source Commitment
Unlike many AI firms offering subscription-based models, DeepSeek has fully open-sourced its models under the MIT license, allowing unrestricted commercial and academic use. This commitment to openness contrasts with the proprietary approaches of some competitors and has been instrumental in its rapid rise in popularity.
DeepSeek has introduced several groundbreaking models since the company’s first launch, including DeepSeek-V3. This model boasts 671 billion parameters and was trained on a dataset of 14.8 trillion tokens over approximately 55 days, costing around $5.58 million. Benchmark tests indicate that DeepSeek-V3 outperforms models like Llama 3.1 and Qwen 2.5 while matching the capabilities of GPT-4o and Claude 3.5 Sonnet. Its architecture employs a mixture of experts with a Multi-head Latent Attention Transformer, containing 256 routed experts and one shared expert, activating 37 billion parameters per token.
The latest model, DeepSeek-R1, released in January 2025, focuses on logical inference, mathematical reasoning, and real-time problem-solving. It was trained using reinforcement learning without supervised fine-tuning, employing group relative policy optimization (GRPO) to enhance reasoning capabilities. This model achieves performance comparable to OpenAI's o1 across various tasks, including mathematics and coding.
One of DeepSeek's most significant contributions is demonstrating that high-performing AI models can be developed with substantially lower costs and resources. For instance, DeepSeek-V3 was trained using approximately 2,000 Nvidia H800 chips, starkly contrasting the 16,000 chips typically used by competitors. This efficiency has prompted a reevaluation of the massive investments in AI infrastructure by leading tech companies.
Awaiting further details
The company has not provided specific details about the nature of the malicious attacks or the anticipated duration of the registration limitations. This situation highlights the challenges rapidly growing AI platforms face in ensuring security and service availability amidst increasing cyber threats.
As the situation evolves, it remains to be seen how DeepSeek will address these security concerns while managing its expanding user base.
