DeepSeek launches V3.1 update, raising prices and targeting next-gen AI agents

Chinese AI startup DeepSeek has released an update to its V3 model, positioning the new version as a faster and more versatile alternative to the company’s earlier R1 system. The update, known as V3.1, has been optimised to run on next-generation Chinese-made AI chips and is being billed as the firm’s first step towards building full-scale AI agents.

In a WeChat post on Thursday, DeepSeek said V3.1 provides quicker answers and enhanced inference capabilities. The company also confirmed that the platform has now gone live on Hugging Face.

The announcement comes months after DeepSeek disrupted Silicon Valley with the powerful yet cost-effective R1, which demonstrated how Chinese firms could make significant advances in artificial intelligence without relying on the most advanced Western semiconductors.

Alongside the product update, DeepSeek introduced a revised pricing plan for the V3 series. The changes, which take effect on 6 September, include higher prices for some services and the removal of discounted evening rates, though fees have been reduced for certain use cases.

Industry observers are awaiting the debut of DeepSeek’s next flagship model, expected to succeed the R1. Reports in Chinese media have suggested the delay stems from founder Liang Wenfeng’s determination to fine-tune the system, while speculation continues about possible training or development challenges.

Meanwhile, major Chinese tech players such as Alibaba and Tencent have continued to accelerate their own AI programmes. Alibaba’s Qwen models, in particular, have gained popularity, intensifying competition in the sector.

On social media, DeepSeek described the V3.1 as “our first step toward the agent era,” highlighting hybrid inference, quicker reasoning, and stronger post-training tool use.