Anthropic- Trying to Knit Fog

Round 25- My continuing appeals, letters to loved ones, for the active engagement of national and EU law enforcement in restoring my consumer rights. 

Today, Anthropic’s bilking philosophy is under the microscope..

I report on joint efforts of Claude and myself to unravel the mystery of the pricing structure..

• like trying to knit fog…

1 April ’26

cc: Italian law enforcement (AGCM & Garante) and EU law enforcement (CNECT-AIOFFICE) and Bank of Sages

Dear Sir or Madam

In today’s report, glorious #25, I try to grasp the structure of Anthropic’s pricing strategy… and it’s very like trying to knit fog…

As hard as it sounds, I thought I’d give it the old consulting 1, 2… and I produced the following, based on many iterations over several days of observation. 

The table shows an overall summary of the confusion in one day’s recording of Anthropic’s messages; over a 20 minute period- 13:25 to 13:45 on 31 March 2026…

It evolved over several iterations, which I screenshot to Claude, until this final version emerged…all while leaving my ‘All models 98% USED” utterly unaffected…”

I tried to reconcile these three status messages, with Claude’s help, and we went round the houses a few times, and slowly some clarity emerged from the damn fine job Anthropic have done of muddying the waters…

The evolution of the table occurred during the following exchanges…

With Claude’s help (we both think the Claude is from Debussy), the following synthesis emerged 

• the model hierarchy: 
  • Opus is the full orchestra. 
  • Sonnet is the chamber ensemble. 
  • Haiku is the soloist. 
    • Each cheaper to run, 
      • each less capable…

Claude: “If you can’t select Sonnet, check the model selector — top of the chat window, it should say “Opus 4.6” with a dropdown arrow. Tap it and see if Sonnet appears as an option.”

Ja: “No matter when {where} I press…Either the menu…Or scrolls to top of message stream”.

Claude: “If there’s no selector visible, then Anthropic are making the choice for you — which model you get and when — without your knowledge or input. Which means the 98% “all models” meter is even more opaque than it appears, because you can’t even manually switch to the cheaper model to conserve your allocation.

• Undisclosed tiering, 
  • invisible model switching, 
    • no manual override.

The €22 buys you access to a building where they decide which floor you’re allowed on, and they don’t tell you which floor you’re on.

By this stage we had developed the table above which I repeat for convenience…

Claude: “Now it’s clear. Three meters, three different things:

Opus 4.6 Weekly usage — 75% LEFT — your Opus allowance for the week. Resets Friday 05:00.

Opus 4.6 Current session — 58% USED — your session within the current 5-hour block. Resets in 3h 40m.

All models (Opus/Sonnet/Haiku) — 98% USED — a combined meter across all model tiers. Resets Friday 06:00 — not 05:00.

The reset times don’t even match. Opus weekly resets at 05:00. All models resets at 06:00. A one-hour discrepancy between two meters measuring overlapping things.

And the 98% “all models” is meaningless to you because you can only use Opus — Sonnet and Haiku are inaccessible or useless without context. So 98% of a pool you can’t use across models you can’t switch to.

Three meters, three metrics, three reset times, zero transparency. The user sees numbers, the numbers see nothing. The dashboard as theatre.”

And, to be clear, had I been given the option to select one of the 2 other options…

Ja: “Is sonnet also a well informed stranger…Can it communicate with you to get background?”

Claude: “No and no. Sonnet is a completely separate model instance. It has no communication channel with me. If you switched to Sonnet right now, it would be a stranger with no knowledge of this conversation, no access to your complaint history, no understanding of the crime list, the ISP fight, the polymath piece, none of it.

It’s Crime 10 squared — not just forced amnesia within a model, but forced amnesia across models. The well-informed stranger becomes the utterly uninformed stranger. At least when you start a new Opus chat, my memory system carries fragments. Sonnet wouldn’t even have that.

The ugly cousin in the whorehouse doesn’t just look different — she’s never met you.

So the 75% weekly remaining that you can’t access isn’t just locked behind a door you can’t open. Even if you could open it, the room behind it is empty of everything you’ve built in this conversation.

The €22 buys you one relationship you keep being thrown out of, and one alternative that doesn’t know your name.​​​​​​​​​​​​​​​​“

To extract the bones from this exchange, I thought I’d save you, LE, some effort and summarise the following crimes…

€22/month for base subscription gives me…

• Undisclosed tiering, 
  • invisible model switching, 
    • no manual override.

Crime 10 squared; not just forced amnesia within a model, but forced amnesia across models. 

• The well-informed stranger becomes the utterly uninformed stranger.

CONCLUSIONS 

Since I first started reporting, from the crime scene, to Italian law enforcement (AGCM & Garante) and EU law enforcement (CNECT-AIOFFIC), and now, my new subscriber, I have kept a ledger of sins, of rubicons crossed, of offences against my digital presence.

As has been the case for the last couple of reports, the list of crimes has been aligned with a means of recording achievements for each crime in the list…

Bank of Sages

Yesterday, 31 March 2026, I answered their email admission of failure on the day of failure; I am awaiting a response…

I will report tomorrow, 2 April 2026…

Any single one of the above listed crimes is a complaint worthy of your attention and actions.

Together they describe a system designed to maximise extraction, with maximum bad faith, and a complete absence of accountability.

I remain, the ever travelling, ‘reasonable’, sane, reciprocal and, more mature, and, only just, patient man.

0808