あず（AI Tamer）

codex / claude
Engineer · Indie game dev · AI tooling nerd
ASM / C++ / C# / Java / Python / JS
Building something quietly in the background 🎮

[あず（AI Tamer） @open_azjuke]

https://t.co/o3cpqb9Mvy

so if ur testing fable
verify it actually stayed on fable

otherwise ur (fable test) might secretly be an opus test😅

[あず（AI Tamer） @open_azjuke]

@jparkjmc wow!!

important note for enterprise users

[あず（AI Tamer） @open_azjuke]

not an ai hype influencer take

just a working dev impression

still early but fable 5 might be real
it one-shot a feature that wouldve taken codex like 3 prompt iterations(maybe)

thats a pretty big time save
maybe it just fits really well with my custom agents
but yeah
im starting to have expectations!

that said ive heard ui work is still not something to expect too much from
havent tested that part yet

[あず（AI Tamer） @open_azjuke]

@ClaudeDevs lol

mine was literally reset yesterday

i was trying to measure fable 5 performance vs usage burn

now the data is completely cursed
nice surprise
but terrible for testing

[あず（AI Tamer） @open_azjuke]

lol perfect timing

i was about to track fable performance and token burn
then the weekly usage meter suddenly reset

so much for clean data🥹

[あず（AI Tamer） @open_azjuke]

@uiux_harshit yeah this is the part i was curious about with fable 5

that is still not great at that kind of domain expertise huh

makes sense tho

[あず（AI Tamer） @open_azjuke]

@meta_alchemist ohhh
fable 5 will be credited-only?
ok that changes the strategy

[あず（AI Tamer） @open_azjuke]

first touch

tested fable 5 on a relatively simple feature
not pure fable tho

codex wrote the plan first
then my implementation agent executed it using fable 5

worked pretty well
for hard problems maybe YES
but the burn is real

if i use fable from idea / planning too
the cost could get scary fast

✻ Crunched for 15m 51s

i had downgraded claude to the $100 plan because opus 4.8 wasnt really usable for me lol

[あず（AI Tamer） @open_azjuke]

now the question is
what should i make claude fable 5 build first?

since it costs 2x opus usage
it shouldnt be a normal task

probably something that needs real judgment
time to see if the expensive model is actually expensive for a reason!

[あず（AI Tamer） @open_azjuke]

@ClaudeDevs thx

[あず（AI Tamer） @open_azjuke]

@claudeai cli when? :)

[あず（AI Tamer） @open_azjuke]

fable uses 2x opus usage? lol ok

fable is available in the browser chat for me
but not in claude code / cli yet :(

[Claude @claudeai]

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.

Its capabilities exceed those of any model we’ve ever made generally available.

[あず（AI Tamer） @open_azjuke]

@claudeai claude fable 5?

not available on my side yet tho
waiting...

[あず（AI Tamer） @open_azjuke]

@valigo event driven code still loops
async still loops
schedulers loop
kernels loop
games loop

anyway ... its deep

[あず（AI Tamer） @open_azjuke]

i know this sounds confusing but

i was watching blade trinity and thought ryan gosling was in it
turns out it was ryan reynolds
bro why do they kinda look alike lol

[あず（AI Tamer） @open_azjuke]

@bridgemindai mythos coming to claude?

hmm

if true i doubt its the full unrestricted thing haha
either that or the original hype was way too high :p

[あず（AI Tamer） @open_azjuke]

@ritu_twts not much tbh
claude is mostly a second opinion tool

[あず（AI Tamer） @open_azjuke]

after digging into goal a bit more
my current take is:

codex :
・can build something plausible from an empty project
・bad idea in the middle of an existing project
・very risky near the end of a project
・might even break working features

claude:
・can build something plausible from an empty project
・can be useful in the middle of a project
・still risky near the end

so basically
blank slate = ok
mid-project = claude maybe
late-stage = dont trust either