Chinese artificial intelligence company DeepSeek has unveiled its latest AI model, DeepSeek V4, which it claims offers world-beating performance and is optimized for domestically produced Chinese semiconductors. The model is available in two versions: V4-Pro and V4-Flash.
Key Features of DeepSeek V4
The new model boasts an ultra-long context window of one million words, allowing it to process vast amounts of text in a single instance. DeepSeek asserts that V4 leads among open-source models in agent capabilities, world knowledge, and reasoning abilities. The V4-Pro variant is said to significantly outperform other open-source models and trails only slightly behind Google's Gemini-Pro-3.1 in world knowledge benchmarks.
Background and Impact
This release follows DeepSeek's R1 model, which last year triggered a trillion-dollar sell-off in Western tech stocks due to its cost-effectiveness and performance that rivaled US tech giants. The new V4 model dramatically improves computational efficiency, processing up to one million tokens and enabling multi-document reasoning. Its launch comes amid ongoing US semiconductor export restrictions targeting China.
- DeepSeek V4 is optimized for Chinese-made chips, highlighting China's push for self-sufficiency in AI technology.
- The model's ultra-long context window supports complex tasks like analyzing entire books or legal documents.
- DeepSeek claims V4-Pro is the most powerful open-source model available, challenging proprietary systems from US firms.
