[pianopatrick]

do you have any data you can share on how many input and output tokens were used in that whole process to fix that bug?

[simonw]

  ~ % uvx agentsview session usage be8850a7-6119-46a0-b5d6-79c7fff5ae2b
  Session:       be8850a7-6119-46a0-b5d6-79c7fff5ae2b
  Agent:         claude
  Output:        68606
  Peak ctx:      113178
  Cost:          ~$12.11 (claude-fable-5, claude-opus-4-8)

[sillysaurusx]

Was the fix worth $12 to you?

[simonw]

I'd have been pretty annoyed if I'd been paying full price, hadn't paid attention and that one prompt (screenshot plus a line of text) had cost me $12!

On the discounted subscription I can tolerate it, it took a small bite out of my daily allowance but not enough that I regret anything.

As an LLM researcher I have no regrets at all because watching it work around the environmental restrictions was fascinating.

[Ucalegon]

How do we know that your pricing or results are normative, given the incentive that any frontier model to juice the pricing/results?

[simonw]

How do you mean?

I'm quoting the API list prices for Fable, at it's $10/million input and $50/million output (and $1/million for cache hits on input).

[Ucalegon]

List price is not paid price and, as Anthropic has shown, initial list price is going to rise.

Do you have pricing, paid by businesses, plesae provide the contracts that prove you assertion that this pricing holds true, or are we supposed to believe you?