ME News update, September 28 (UTC+8): DeepSeek officially announced that the underlying model of deepseek-chat has been upgraded to DeepSeek-V3-0324. The new version shows a significant improvement in reasoning ability, with benchmark scores such as MMLU-Pro, GPQA, AIME, and LiveCodeBench all seeing substantial gains (e.g., AIME increased by nearly 20 points). In addition, the accuracy and aesthetics of web front-end code generation have been optimized; the Chinese writing style has been aligned with R1; Function Calling accuracy has improved and previously encountered issues have been fixed. 🔗 Read the original via AIHOT · https://aihot.news/items/cfz5ubf0pr6evxssrvx0j0vtl (Source: ME)