@heyandras Used a 3060 12GB here. It's the same architecture but with more VRAM. For the 3000 series, it's decent for running image generation models, tbh.
Before adding another “token saver”, measure what you already have,
your cache hit rate.
Here @pidotdev extension that measures from your actual workflow/sesions history.
No synthetic benchmark, No “trust me bro” test.
actually someone literally sued Anthropic over this in June. the lawsuit claims that Claude Max’s usage limit marketing is misleading
based on Anthropic’s own weekly estimates in the lawsuit:
5x plan: ~3.5x Pro weekly
20x plan: ~6x Pro weekly
you pay 2x: get ~1.7x more weekly usage
and this is before Fable.
the lawsuit claims Anthropic sells “5x / 20x more usage” but the weekly limits aren’t meaningfully explained in the purchase flow.
*the case is still ongoing so these are allegations, not a court finding yet.
@badlogicgames Do TreeSitter / AST based search tools help reduce context pollution & code rot in agents, or can they still amplify it? Worth using?