Chase AI
Region: US
Thursday 06 August 2026 06:13:47 GMT
9224
348
9
77
Music
Download
Comments
JoeyC.ai :
I’ve been using /clear frequently like you taught in your CC masterclass but I also have been using Fable 5 as the architect and sonnet 5 or opus 5 as the builder. That works great!
2026-08-06 12:18:05
0
ImTokkingIt :
Bro unironically suggested using Sonnet 😂
2026-08-06 10:43:46
0
Luke Rhodes :
Subagents default to 5m cache too, costs less though
2026-08-06 10:43:18
0
ZenPanda :
Great explanation! You presented this token consumption in a very useful way!
2026-08-06 11:07:14
0
Michael Langdon :
So practically speaking, do you want to compact at the start or end of a conversation? Or does it make no difference?
2026-08-06 07:26:05
0
Luke Rhodes :
Caveman works well but ponytail had too big an impact on my benchmark results; 25 points on a 110 frontier task bench similar to deepswe. Input tokens is where all of the cost is despite being cheaper than output - for most people. “Headroom” with ~17% saving on input is a huge help. Auto compacting at more appropriate times and auto-keeping the 1h cache alive if you’re using the computer can help another 2-3%.
2026-08-06 10:49:05
0
somaforthepeople :
Ponytail is the bomb. I concur. You skipped lean ctx. How come?
2026-08-06 06:49:17
0
WelldadLeuven :
no.1 should be subscribe to chatgpt codex
2026-08-06 10:13:53
0
KreyolGenius :
😂😂😂
2026-08-06 06:20:15
0
To see more videos from user @chase_ai_, please go to the Tikwm
homepage.
