DeepSeek v4.1 Flash Is Now Our Best Hacking Model

DeepSeek V4.1 Flash achieved a perfect score on an AI hacking benchmark, successfully exploiting all 11 vulnerable targets while keeping 4 fixed targets secure, costing only $4.65 through token caching. A detailed audit revealed the model not only followed six planned attack paths but also discovered five unexpected exploitation routes that the original scoring system didn't distinguish. The benchmark review demonstrated both DeepSeek's strong hacking capabilities and the need for more rigorous evaluation criteria that examine both outcomes and the methods used to achieve them.