DeepSeek V4 is a display of Huawei AI chip's capabilities
5 hours ago
- #Artificial Intelligence
- #Chip Independence
- #Open Source
- DeepSeek V4 is a new, efficient open-source model with high performance at a low cost, offering two versions: V4-Pro for coding/complex tasks and V4-Flash for speed/affordability.
- The model features a long context window of 1 million tokens, achieved through architectural innovations that reduce computing power and memory usage, making it cheaper for handling large texts.
- V4 is optimized for Chinese chips like Huawei's Ascend, marking a step toward reducing dependence on Nvidia and aligning with China's self-reliance goals in AI infrastructure.
- The release follows DeepSeek's rise from a little-known team to a leading AI company after R1, though V4 faces scrutiny over personnel changes and government pressures.