DeepSeek AI has released an updated version of its DeepSeek-V2.5 model, named DeepSeek-V2.5-1210. This new iteration boasts significant enhancements across various capabilities. Mathematical performance has seen a notable improvement, with the MATH-500 benchmark score rising from 74.8% to 82.8%. Coding accuracy has also increased, with the LiveCodebench benchmark score going from 29.2% to 34.38%. Additionally, the model now supports online search, enhancing its ability to provide up-to-date information. Writing and reasoning capabilities have been improved based on internal tests, and file upload functionalities have been optimized. The release of DeepSeek-V2.5-1210 marks the final installment in the V2 series, with the company setting its sights on the upcoming V3 release.