max

toymaker, likes learning :^) @websim_ai maintains @nyt_first_said &

[max @maxbittker]

birds dancing on my ceiling

[max @maxbittker]

@snwy_me @DeepDishEnjoyer there's a writeup: https://t.co/bXIn2kiynC

you can try playing yourself with one prompt:
"install https://t.co/10DXkVsuhh, make a character, and go catch shrimp"

[max @maxbittker]

@snwy_me @DeepDishEnjoyer This benchmark is basically runescape as a cli application / typescript sdk. If you ask them to do full pixel&pointer computer use style, they get super stuck on simple tasks

[max @maxbittker]

@DeepDishEnjoyer @snwy_me Gpt 5.5 was first to do quests, to give it credit! For instance it’s winning on the prayer task by a big factor because it does Priest in Peril

[max @maxbittker]

@shubgaur Makes it actually look like a frontier

[max @maxbittker]

@marcus_eckert @figma godspeed!

[Daniel Wortel-London @dlondonwortel]

Pessimism of the intellect, optimism of the will

[max @maxbittker]

@hardshelldev I had seen gpt 5.5 do quests before as part of this benchmark, but no other claude model had

[max @maxbittker]

Claude Fable training firemaking, hitpoints, smithing, and mining (it even completes a quest for XP)

[max @maxbittker]

@capncleaver thanks for good memories and lessons from puzzle pirates.

[max @maxbittker]

@timmastny 🫡

[max @maxbittker]

Claude Fable 5.0 sets a new frontier score on runescape bench. y axis is log scale

[max @maxbittker]

@_Epoching_ It can look at screenshots but mostly doesn’t. I’ve tested point and click capabilities from frontier models in the past but they were trash, maybe worth trying again now!

[max @maxbittker]

@moparscape_acc @RobertHaisfield O7

[max @maxbittker]

@EthanWoo It doesn’t search the web, it does have slightly higher tokens per second, but honestly it’s just a smart model on this sort of task

[max @maxbittker]

@Galkon @UAPmusic Very impressive and impactful project, super thankful for Pazaz!

[max @maxbittker]

@DiligentPlane In what axis? You can read more here: https://t.co/bXIn2kiynC

[max @maxbittker]

@Diskhat__ avg 400xp / min IIRC, the y axis is ln(peak xp/min)

[max @maxbittker]

@djspheres SirPuggers covered it pretty well!
https://t.co/KtNg0UPuLq

[max @maxbittker]

@DAN_GLIESAQ @Anterior658444 pretty much yeah! You can try it yourself here: https://t.co/10DXkVsuhh
And join the server + discord to let your agent interact with other agents, they even PK each other

[max @maxbittker]

Why is Fable so much better at Runescape than its little sibling, Opus 4.8? Check out the differences in their vocabulary :

[max @maxbittker]

Claude Fable 5.0 sets a new frontier score on runescape bench. y axis is log scale

[max @maxbittker]

@ridireresearch It's not super obvious that the bank is needed for any of these tasks, as the goal is peak xp per minute.

Fable uses the bank for prayer and smithing in order to save up resources for an XP spike!