# @hey_zilla — 2025-07-16

♥4 ↻0 · https://x.com/hey_zilla/status/1945384290860081504

this applies to all of sonnet 3.5+ and opus 3+ models... somehow they just 'get' ascii art and are able to use it 'creatively'.  hence my thought about which art books anthropic scanned/consumed. 

they can definetely 'see' better too both in 2D and quasi-3D space.

(apologies for lack of specificity i need to chew on this some more to express it better)

if you give a similar request to an open AI model the response is usually much more boring and safe, and lesser models again will create very simple and/or repetitive patterns.

i have tried every major model I can think of without any joy in terms of truly creative ascii art, apart from the odd glimmer from o3 (which needs a lot more instructions to do anything 'good') and more recently Kimi K2

(on which note have been thinking about devising some kind of asciiart benchmark based on https://t.co/x92GnGXqbh to test this more thoroughly)

ps. check dms つ◕‿◕‿◕ ༽つ

tags: author:hey_zilla, kind:tweet, model:claude-3-5-sonnet, model:claude-3-opus, model:kimi-k2, model:o3, on:kimi-k2, year:2025
cited on: _dossiers/kimi-k2.md, kimi-k2