DeepSeek Just Re-Trained the Same Model and Gained 145 Points. The Weights Are MIT.

DeepSeek V4 Flash-0731 Impresses Early Testers with Strong Performance-to-Cost Ratio Ahead of Pro Release

Igeekphone News August 1, 2026 – Chinese AI startup DeepSeek has officially released DeepSeek V4 Flash-0731, the first public version of its next-generation V4 model family. While the company has positioned the model as a lightweight offering, early benchmarks and community testing suggest it delivers performance that rivals significantly larger AI models, particularly when evaluated against its API pricing. The release comes just ahead of the a…

DeepSeek Just Re-Trained the Same Model and Gained 145 Points. The Weights Are MIT.

By Evan Vega DeepSeek released a new checkpoint of V4-Flash on 31 July 2026. The architecture did not change — still two hundred eighty billion parameters total, thirteen billion active per token — and neither did the price, at fourteen cents per million input tokens and twenty-eight cents per million output. What changed was the post-training. On Arena.ai’s Frontend Code Arena, the result was a jump of a hundred and forty-five points. The more …

DeepSeek's Cheap Model Just Beat Its Own Flagship on Nine Benchmarks

DeepSeek moved its V4-Flash model out of preview and into official public beta on July 31, 2026, and the upgraded version now outscores the company’s own larger, more expensive V4-Pro-Preview model across every agent benchmark DeepSeek published. The catch: it’s the same architecture as before. DeepSeek only retrained it. V4-Flash-0731 keeps the identical 284-billion-parameter mixture-of-experts design and 13-billion active parameters as the pre…