DeepSeek news

DeepSeek auto-feed stories: new V4 family models, API pricing, and privacy risks. Important ones - with a date and proof.

1. November 6

   DeepSeek nears OpenAI's closed model with 72.8% of objectives achieved in a hacking test

   In the Oct 23 CyberOffense test, the downloadable DeepSeek V4.1 Flash 1017 achieved 72.8% of objectives, compared with 80.8% for the closed GPT-6 Sol.

2. November 5

   Super-Agent: DeepSeek's AI agent takes chat tasks for 1 GitHub repository

   On Oct 31, DeepSeek released the source code for Super-Agent, a research prototype for 1 GitHub.com repository. The agent takes tasks from Feishu or follows team rules and returns changes through PRs.