Skip to content 
Key takeaways
- DeepSeek V3 Launch: A powerful open AI model, released with a permissive license for commercial and personal use.
- Performance Leader: Outperforms leading models like Meta’s Llama 3.1 and OpenAI’s GPT-4o in coding tasks and benchmarks.
- Massive Scale: Trained on 14.8 trillion tokens with 671 billion parameters, setting new standards in AI development.
- Cost-Efficient Training: Developed in two months using Nvidia H800 GPUs for just $5.5 million.
- Regulatory Limits: Responses align with Chinese internet standards, restricting answers on sensitive topics.
- Visionary Backing: Funded by High-Flyer Capital, aiming for open-source AI to compete with closed models.
Revolutionizing AI Development with Massive Scale and Accessibility
China’s AI firm DeepSeek has launched DeepSeek V3, a cutting-edge AI model that stands out as one of the most powerful “open” models to date. Released under a permissive license, DeepSeek V3 empowers developers to download, modify, and apply it for a wide range of uses, including commercial applications.
Key Features and Performance Highlights
DeepSeek V3 is designed for diverse text-based tasks, including coding, translation, essay composition, and email generation from descriptive prompts. According to DeepSeek’s internal benchmarks, it surpasses both open-source and proprietary models, such as Meta’s Llama 3.1 405B, OpenAI’s GPT-4o, and Alibaba’s Qwen 2.5 72B.
In coding challenges hosted on the competitive platform Codeforces, DeepSeek V3 has demonstrated superior capabilities, including its ability to write and integrate new code seamlessly, as measured by the Aider Polyglot test.
Massive Training and Parameter Scale
DeepSeek V3’s success stems from its monumental training dataset of 14.8 trillion tokens, equating to approximately 11.1 trillion words. Its architecture boasts 671 billion parameters, scaling up to 685 billion parameters on platforms like Hugging Face. This parameter count is 1.6 times larger than that of Meta’s Llama 3.1 405B.
Although larger models typically demand robust hardware, DeepSeek has managed remarkable efficiency, training the model in just two months using Nvidia H800 GPUs. The entire project cost a modest $5.5 million—a fraction of the expense incurred by competitors like OpenAI’s GPT-4.
Challenges and Limitations
While DeepSeek V3’s technical achievements are undeniable, certain limitations exist. The model’s responses are regulated to align with China’s internet standards, adhering to “core socialist values.” This results in restricted responses to politically sensitive topics, such as Tiananmen Square or speculative questions about Chinese leadership.
DeepSeek’s Ambitious Vision
DeepSeek operates under the aegis of High-Flyer Capital Management, a Chinese quantitative hedge fund leveraging AI for trading strategies. High-Flyer’s advanced infrastructure includes a server cluster equipped with 10,000 Nvidia A100 GPUs, valued at $138 million.
High-Flyer’s founder, Liang Wenfeng, envisions a future where open-source AI will close the gap with proprietary systems, calling closed models like OpenAI’s a “temporary moat.” The recent launch of DeepSeek-R1, their competitor to OpenAI’s reasoning model, reflects this ambition.
DeepSeek V3 marks a milestone in AI innovation, blending accessibility with world-class performance. Despite geopolitical and regulatory constraints, its launch sets a new benchmark for open AI models and underscores China’s rapid advancements in the field.
More from Earn Hustles Insider:
Connect with us on LinkedIn to expand your network and grow together: Earn Hustles Insider.
