GoKawiilDeepSeek V4.1 Flash tops AI hacking benchmark, cracks 11 of 11 targets for $4.65
DeepSeek V4.1 Flash achieved code execution on all 11 vulnerable systems in an AI hacking benchmark while leaving four patched systems untouched, at a total cost of just $4.65 for accepted runs. A manual review found the model discovered five novel attack paths beyond the six expected solutions, including a faster exploit against Grafana that bypassed the intended vulnerability entirely.
GoKawiil's interpretation of the reporting above, not reported fact.
The extremely low cost, driven by heavy token caching, shows that sophisticated automated exploitation is becoming cheap and accessible, not just technically feasible. The discovery of unscripted attack routes also exposes weaknesses in how security benchmarks score success, since scoring systems failed to distinguish planned exploits from entirely different vulnerabilities the model found on its own.
- DeepSeek V4.1 Flash exploited all 11 vulnerable targets and avoided all 4 secure ones in the benchmark.
- Total cost for accepted runs was $4.65, aided by 266.2 million cached input tokens out of 268.3 million.
- The model found 5 novel exploit paths not accounted for in the benchmark's original scoring, including a faster Grafana attack.
Source: enclave.ai — Yanir Tsarimi, 2026-09-16
Published there as: “DeepSeek v4.1 Flash Is Now Our Best Hacking Model”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.