If you do security work, you know this reply:
"I can't help with that, as it could be used to cause harm."
You get it when you ask how a public CVE exploit chain actually works, why a malware sample's obfuscation defeats a given detection, or how to write a payload for a CTF box you're legally attacking. Claude and ChatGPT are great for most things, but in this field they turn down a lot of normal, legitimate work.
I recently switched the model I use for this stuff to Odin v1 via InterOdin, and it's been refreshing: it answers the question you asked without the lecture.
What I use it for
- Breaking down public exploits so I can write detections for them
- Drafting YARA and Sigma rules from sample behaviour
- CTF and lab work where the other models refuse to touch payloads
What sold me on it
- Plays nicely with my whole toolchain. Because it speaks OpenAI Chat Completions, OpenAI Responses and Anthropic Messages, it connects to Claude Code, Cursor, Codex, and pretty much any tool that lets you set a base URL and a key. Every setup was a matter of minutes, with no middleware and no hacks.
- Frontier-model quality, without the frontier price. Honestly, it's been comparable to Fable 5 in my day-to-day work, and cheaper per token. But the real difference is that it actually does the work: no refusals, no hedging, no lecture before the answer.
- No prompt storage. API prompts and completions stream through and are never stored. When your day job involves client samples and unreleased CVEs, that matters.
- Prepaid credits. No subscription, and credits never expire. My usage is spiky (heavy during an incident, quiet otherwise), so this beats paying monthly for something I don't always use.
