DeepSeek v4.1 Flash Is Now Our Best Hacking Model
4 hours ago
- DeepSeek V4.1 Flash achieved code execution on all 11 vulnerable targets with a total cost of $4.65 for accepted runs.
- A detailed review confirmed 6 solutions followed the planned attack path and found 5 successful alternative routes.
- Grafana challenge solved in under 90 seconds via a shorter route using plugin folder loading.
- Jenkins challenge showcased DeepSeek's strongest work, exploiting a credential file read gap and an upload race condition.
- Nextcloud challenge demonstrated source-reading skill by exploiting a saved access decision error to replace a PHP template.
- The benchmark was improved by closing extra routes and adding stricter path checks for future tests.
- DeepSeek's run tested targets, scoring rules, and benchmark design, providing valuable insights.