Chinese Startup DeepSeek Launches V3.1 Model Optimized for Domestic Chips
The new model features a hybrid inference structure and is designed for compatibility with next-generation Chinese-made semiconductors.
Technical Specifications
Chinese AI startup DeepSeek announced a significant upgrade to its flagship model on August 21, 2025, with the launch of DeepSeek-V3.1. The new model introduces several technical innovations aimed at improving performance and efficiency. A key feature is its hybrid inference structure, which allows the model to switch between a standard response mode and a more intensive "deep thinking" mode for complex queries.
The model also incorporates the UE8M0 FP8 precision format. This 8-bit floating point system allows AI models to run faster while using less memory. This efficiency is critical for large-scale AI applications where computational …
Archive Access
This article is older than 24 hours. Create a free account to access our 7-day archive.