DeepSeek v4.1 Flash Is Now Our Best Hacking Model

DeepSeek V4.1 Flash achieved code execution on all eleven targets in an AI hacking benchmark, completing the test runs for a low cost due to heavy token caching. The review confirmed multiple successful attack routes and highlighted areas for stricter benchmark checks, though independent verification of the general performance claims is not included.

Cover image for DeepSeek v4.1 Flash Is Now Our Best Hacking Model