China’s DeepSeek Gets a Model Upgrade with V2.5-1210

DeepSeek, a Chinese AI research lab backed by High-Flyer Capital Management, has launched V2.5-1210, the final model in its V2 series. The model introduces Internet Search capabilities for real-time answers and excels in tasks like math, coding, writing, and roleplay.

The model is accessible on chat.deepseek.com. Users can toggle the Internet Search feature on the website for real-time responses or integrate the model via Hugging Face. 

Developers can explore its capabilities and build applications using the open-source release. The release aims to meet diverse user needs, enhance productivity, and provide a versatile AI tool for work and life applications. 

With this tool, the company is directly competing with the likes of OpenAI’s ChatGPT and Perplexity’s Search engine. However, users on X claim that Deepseek is still in its early stages and lacks widgets. However, it is capable of managing basic tasks, such as real-time search-based prompting and reviewing 50 different sources when querying for news.

“As V2 closes, it’s not the end—it’s the beginning of something greater. DeepSeek is already working on next-gen foundation models, and the DeepSeek V3 series will be released in the future to push boundaries even further. Stay tuned!,” read their post on X. 

Last month, the company released DeepSeek-R1-Lite-Preview, a reasoning AI model rivaling OpenAI’s o1. It matches OpenAI’s performance on benchmarks like AIME and MATH, offering step-by-step “chain-of-thought” reasoning for transparency. 

The model improves with longer reasoning outputs, challenging traditional AI scaling laws by using additional processing time for complex tasks. Available via DeepSeek Chat with a 50-message daily limit, it faces regulatory restrictions on politically sensitive topics. 

DeepSeek plans to release open-source R1 models, increasing competition with Chinese tech giants like ByteDance, Alibaba, and Baidu. Alibaba’s rival Qwen2.5-Turbo supports massive context lengths.