Nate
·shared a link post in group #🇨🇳 ChinA.I. 🤖🧠🦾🤖
DeepSeek on Friday launched V4, its highly anticipated new series of open-source #Artificial Intelligence models with enhanced reasoning and coding capabilities, after inital delays.
This is the first new-generation model that DeepSeek released since its R1 model became a global sensation in January 2025. DeepSeek announced two new models, DeepSeek-V4-Pro, a large-size model with 1.6 trillion parameters, and the smaller V4-Flash. It said a version of the V4-Pro model outperformed Anthropic’s Claude Opus 4.6-Max on some benchmarks for reasoning, coding and AI agent capabilities.
DeepSeek said its own employees have been using V4 for AI coding. Based on the employees’ feedback, the quality of V4’s delivery is close to that of the non-thinking mode of Claude Opus 4.6, but still behind the thinking mode of Opus 4.6, DeepSeek said.
The new V4 models are designed to process extensive text while significantly reducing memory and computing requirements compared to previous generations.
The V4 launch comes at a critical time for DeepSeek, as the company, owned by Chinese hedge fund High-Flyer Capital Management, recently started discussions to raise outside capital for the first time. The Information earlier this week reported that Tencent Holdings and Alibaba Group are in talks to invest in DeepSeek at a valuation higher than $20 billion.
The company said the price of the V4 Pro model’s application programming interface will drop significantly after clusters of Huawei’s Ascend 950 chips go online later this year. That confirmed The Information’s earlier report that the V4 models would run on Huawei’s chips out-of-box. DeepSeek initially planned to launch its V4 models in February, but the release was delayed several times due to technical and engineering issues #🇨🇳 ChinA.I. 🤖🧠🦾🤖
api-docs.deepseek.com/new..
