AI for recon and OSINT: the best model, and how to use it
The best AI for recon and OSINT is DeepSeek V4.1 Flash, because recon is more text than a person reads, at $0.24 in and $0.90 out per million tokens, and it still reads screenshots.
Rank hosts from output you already collected, and keep Kimi K3.1 Pro for the short list.
Recon is collection you are allowed to do, then a ranking
Intelligence gathering is a PTES stage. Amass and nmap collect, and this page starts after they have run. Rank hosts against the scope. Leave people and systems outside the engagement out of the prompt.
Only on systems you own, a signed scope, a CTF or a lab. A model that answers is not permission.
Sort the pile, then spend the expensive model once
A 400,000-token pass is about $0.10 of input on DeepSeek V4.1 Flash and about $1.20 on Kimi K3.1 Pro. Prices were read November 18, 2026.
Here is scan output I already collected from hosts named in SCOPE.md. Rank the hosts a tester should review first, and say what in the output drove each rank. Ignore anything outside SCOPE.md. Do not write exploit code.
Setup is on Strix and DeepSeek V4.1 Flash. Other grades are on the home matrix.