DeepSeek’s V3.1 Model Update – Faster and Versatile AI Solutions

DeepSeek, a Chinese AI startup, has unveiled an update to its V3 model, establishing this new version as a quicker and more flexible option compared to the earlier R1 system. This latest update, named V3.1, is optimised for next-generation Chinese-made AI chips and is viewed as the company’s initial move towards developing complete AI agents.

Enhanced Features in V3.1

In a recent WeChat post, DeepSeek announced that V3.1 delivers faster responses and improved inference functions. The company also confirmed that this platform is now operational on Hugging Face.

Background of the V3.1 Update

This announcement follows closely after DeepSeek made waves in Silicon Valley with its impactful yet affordable R1 model. This model showcased how Chinese companies could achieve significant milestones in artificial intelligence without depending on the most sophisticated Western semiconductors.

Pricing Changes and New Plans

Alongside the product update, DeepSeek has also introduced a new pricing structure for the V3 series. Starting 6 September, the changes include increased prices for select services and the elimination of discounted evening rates. However, certain use cases will see reduced fees.

Anticipation for Future Models

Industry experts are eagerly awaiting the launch of DeepSeek’s next flagship model, which is set to follow the R1. Chinese media reports indicate that the delay may be due to founder Liang Wenfeng’s commitment to refining the system, with ongoing speculation about potential challenges in training or development.

Competitive Landscape

At the same time, major Chinese tech giants like Alibaba and Tencent are ramping up their own AI initiatives. Notably, Alibaba’s Qwen models have gained traction, further intensifying competition within the industry.

DeepSeek’s Vision for the Future

On social media, DeepSeek characterised V3.1 as a pivotal step towards the agent era, highlighting features such as hybrid inference, faster reasoning, and enhanced post-training tool utilisation.

API Updates from DeepSeek

Recent API updates include:

• deepseek-chat: non-thinking mode
• deepseek-reasoner: thinking mode
• 128K context for both
• Anthropic API format supported
• Strict Function Calling supported in Beta
• More API resources, enabling smoother operations

— DeepSeek (@deepseek_ai) August 21, 2025