DeepSeek shows off new agentic AI model that surpasses R1 in key areas

[DeepSeek Shakes Up The Global AI Market]
Leon Neal/Getty Images News

On Thursday, Chinese AI startup DeepSeek (DEEPSEEK [https://seekingalpha.com/symbol/DEEPSEEK]) officially launched its updated DeepSeek-V3.1 AI model, which surpasses its R1 model on key benchmarks.

The company unveiled [https://seekingalpha.com/news/4486964-deepseek-unveils-update-to-v3-model] V3.1 earlier this week.

DeepSeek said the new model has hybrid inference, as the model supports both thinking mode and non-thinking mode.

"Faster thinking: DeepSeek-V3.1-Think reaches answers in less time vs. DeepSeek-R1-0528," said the company in a post [https://x.com/deepseek_ai/status/1958417062008918312] on X on Thursday.

The company noted [https://mp.weixin.qq.com/s/WUbmBSapVyvxZe6HobD5Qw] that the model also has stronger agent capabilities, adding that through post-training optimization, the new model has significantly improved its performance in tool usage and agent tasks.

On Thursday, DeepSeek also outlined [https://x.com/deepseek_ai/status/1958417076114334160] a new pricing [https://api-docs.deepseek.com/quick_start/pricing/] plan, raising prices in some cases and discontinuing discounted evening rates. The company has reduced fees for certain use cases.

DeepSeek shook [https://seekingalpha.com/news/4399000-nvidia-broadcom-lead-tech-stocks-lower-amid-deepseek-ai-impact#hasComeFromMpArticle] the U.S. tech sector in January after it unveiled [https://seekingalpha.com/news/4398699-chinese-ai-lab-deepseek-rattles-silicon-valley#hasComeFromMpArticle] its R1 large language model, or LLM, and detailed in a paper how LLMs can be built on tight budgets and improved without human supervision.

The company, reportedly [https://seekingalpha.com/news/4485119-deepseek-delays-launch-of-new-model-amid-huawei-chip-setback-ft], has pushed back the launch of its new AI model R2 after hitting obstacles in training it with chips from Huawei Technologies.

MORE ON DEEPSEEK

* DeepSeek unveils update to V3 model [https://seekingalpha.com/news/4486964-deepseek-unveils-update-to-v3-model]
* DeepSeek delays launch of new model amid Huawei chip setback: FT [https://seekingalpha.com/news/4485119-deepseek-delays-launch-of-new-model-amid-huawei-chip-setback-ft]
* Financial information for Hangzhou DeepSeek Artificial Intelligence Co., Ltd. [https://seekingalpha.com/symbol/DEEPSEEK/income-statement]