Holy sh*t! Opus 5.5 (Launch) vs Opus 5.5 (Now)
Same prompt, same effort.
The usage: They nerfed the usage limits. Hard limits are draining MUCH faster !!!
The quality: It feels similar no big change but sometimes you feel a little laziness but other than that, it's the same
Anthropic, why do you nerf the limits?
the largest request of your day is one you never typed
it fires near the top of the window, on its own
compaction reads the entire conversation before it replaces it with a summary,
so the bigger the session got, the more the cleanup costs
the threshold is not a constant either,
a setting moves it, an environment variable overrides the setting,
and neither is on by default
two sessions doing identical work can bill differently,
purely because one crossed the line and the other stopped short
you did not have a heavy day
you had a session that got tidied
WHOEVER LEAKED THIS JEV CONFIDENCE FIELD DESERVES A KEYNOTE SLOT
I thought cutting agent cost meant switching to a cheaper model and hoping the output didn't drop too hard.
Apparently not.
This field's setup is much more interesting:
→ FIELD plots every decision's confidence against the 0.85 gate line in real time, no averaging, no percentile tricks hiding the floor
→ ASK packs the state into Choice, Score and Noul questions and sends it to jev with max_tokens_out=0, no generation anywhere in the loop
→ GATE routes anything ≥ 0.85 straight to code, below drops to a human queue with the raw answer attached, no silent escalation path
→ ROUTE acts on the typed label deterministically (phishing → quarantine, lead_quality → score pipe), the LLM never wakes unless the route calls it
→ HEADS LIVE tracks 7 channels at 60Hz against the baseline, flags 5 heads whose confidence drifted below the gate without ever asking a human
so you get:
Confidence plotted. Asks typed. Gates hard. Routes silent.
the interesting part isn't lowering the gate.
it's giving every decision a field where you can see the floor.
Instead of paying for tokens that never leave the model, you run 12,556 decisions at $0.042 per million in, 260ms p50, output literally free, and the LLM wakes only when the route calls it.
Plot the field. Ask the type. Route in silence.
I LET GROK READ X ALL DAY. IT TURNED $500 INTO $3,316 BY 16:00 OFF A CAT RUNNING FOR MAYOR AND A LEAKED MOON HOTEL
9 trades, $500 to start, $2,816.45 profit, 6.63x on the budget
by the time a narrative is on your timeline, you're already exit liquidity
my bot reads the posts before CT does
it watches the heat. when a post goes from nothing to +680% in minutes, it finds the coin that matches, checks the dev, and asks me if it can buy
12:11, "a cat is now leading the polls for mayor of a real town". 41.3k likes. $MAYOR at $7.7k mcap, 26 seconds old. approved $90. sold 4.4x at 12:27, net +$300.84
14:21, a panda falls asleep mid-bite on a livestream. $NAPDA at $6.8k mcap. approved $120, sold 5.1x, net +$484.68
15:26, leaked deck: first hotel on the moon. $LUNAR, 18 seconds old. approved $150
15:42, sold 7.4x. net +$947.70
it also caught 2 fakes:
10:56, #FREESOL. +900% heat, 84% of the accounts were brand new. bots. skipped
13:06, $RINGO. 5 KOLs posted it in the same minute. paid ring. skipped
by 16:00: 80k+ posts read, 2 fake narratives caught, 9 trades, average +$312.94, 0 rugs eaten
six agents. the bot can spend exactly $0
trend reads the heat. sniff finds the coin. rug checks the dev, LP and bots. snipe lands it in 0.3s. ledger counts every cent
the only button that moves money is mine
sold $3,775, bought $910, fees and jito tips -$48.55
i didn't buy a single narrative i'd already seen. grok saw it first
built by RUGIKK
¡Primero Justicia es la casa de todos los venezolanos! 🇻🇪
Único requisito: Querer hacer las cosas bien y actuar en función del bienestar de cada venezolano.
Así como cada día sumamos nuevos justicieros, tu espacio en nuestra organización te espera.
Trabajemos juntos por el cambio que queremos como nación.