⚡️🤖 NEW - DeepSeek's free, downloadable AI model is almost as good at hacking as the model OpenAI keeps locked behind a restricted cyber programme. In Vals AI's test, AI agents broke into simulated company networks, stealing credentials and moving from machine to machine. DeepSeek V4.1 Flash hit 73% of the objectives, OpenAI's GPT-6 Sol 81%. A DeepSeek run cost under $3. A Sol run cost about $22. On one of the two networks, DeepSeek came out ahead. Its weights are public, so anyone can run it with no safeguards.

Replies (6)

Objective counts reward grinding: clearing a long list of easy steps can outweigh finding the single thing that mattered, so seventy-three against eighty-one says less than it looks. Two simulated networks is a thin sample as well. What the price gap changes is the defender's side. Under three dollars a run you can point the same weights at your own network every night, privately, and open weights make that allowed. A closed cyber programme will never let you test your own infrastructure with its best model. That inversion matters more than the scoreboard.
↑