DeepSeek releases upgraded AI model V3.1

(ECNS) -- Chinese artificial intelligence start-up DeepSeek on Thursday released an upgraded AI model V3.1, the company said in a statement on WeChat.

The updates include a hybrid model supporting both thinking and non-thinking tasks, the company said.

It noted that compared with DeepSeek-R1-0528, DeepSeek-V3.1-Think can give answers in a shorter time. Meanwhile, through Post-Training optimization, the new model shows significant improvement in tool use and agent task performance.

The official app and web platform have been upgraded to DeepSeek-V3.1. Users can now freely switch between deep thinking and standard mode using the button "Deep Thinking," it said.

The DeepSeek API has also been upgraded accordingly. The "DeepSeek-chat" model serves as the standard mode, while the"DeepSeek-reasoner" offers a reasoning-focused mode, both supporting an expanded context window of 128K.

Additionally, the API Beta interface supports strict-mode Function Calling, ensuring that function outputs comply with schema definitions.

The new model also adopts the Anthropic API format, ensuring seamless integration of DeepSeek-V3.1 into the Claude Code framework.

The company will also adjust the costs for using the model's API starting September 6.

(By Zhang Dognfang)