DeepSeek-V4 has entered small-scale gray testing, with two versions included: DeepSeek-V4-Pro and DeepSeek-V4-Flash. According to ChainCatcher, sources familiar with the matter said the official version is now expected to launch by the end of this month, later than previously planned.
The new release is said to focus on improving reasoning efficiency, response speed, and agent capabilities. Developers participating in the test said V4's overall performance is comparable to Anthropic Opus 4.8, its code generation is close to OpenAI GPT-5.6 Sol, and it shows better performance in 3D modeling, SVG image generation, and small-game development.
DeepSeek-V4 has also completed CUDA-free adaptation and is expected to be compatible with multiple domestic AI chips on launch day. It will use peak-and-valley API pricing, with peak-hour rates set at twice the off-peak price.
DeepSeek-V4 Begins Small-Scale Gray Testing Ahead of Expected Late-Month Release
2026-07-21 13:43:57
Disclaimer:
1. The information provided does not constitute investment advice. Investors should make independent decisions and bear all risks themselves.
2. The copyright of this content belongs to the original author. The views expressed herein are solely those of the author and do not represent the stance or position of this website.
