While building Nostrautica, I am testing models with spare tokens to do security, reliability and UX audits.
Some learnings:
- Nostrautica was designed by Fable and implemented by Opus. Pretty good, worked pretty much out of the box, impressive.
- Kimi K3 suggested meaningful improvements, found security bugs, etc. It is a good model, although it was pretty "expensive" (eats many tokens)
- Yesterday I ran a full re-audit (after fixing those bugs) with GPT5.6 Sol, it cost only $20 (limit was $35) and it found even more bugs, including pretty serious security bugs. Race conditions, replay attacks, etc.
My take-away: There is no "this model is best". And if you want to use AI for audits, run more models, they will find different things. And both Kimi K3 and GPT5.6 Sol are great models, GPT5.6 Sol I found to be very good in debugging and finding bugs (much better than Opus - and Fable will chicken out soon).
Nostrautica is going pretty well, we'll use it for our first event in a few days. For those who don't know what it is, it is next-gen decentralized event organizer/attendee app that runs on Nostr, helps match people. If you are on Nostr already, it uses what it knows about you to find you good matches to meet in person. If you are not yet on Nostr, you will have to feed it info about you, but you will end up with Nostr identity - filled profile, already following people, private conversations and contacts. You can then stop using Nostrautica and continue on Nostr-proper.
Nostrautica
Nostrautica — chart a course to the people worth meeting
Nostrautica pairs AI matchmaking with a real, portable Nostr identity, so every attendee knows who to meet, and why. For a normal conference with l...
Here is a test event if you want to try it out (we need to manually approve you, so it will take a bit):