groq’s kimi-k2.1 at 400 TPS with Claude Code is unbelievable.
Spent last 6 hours or so figuring out how to set up the groq + kimi to anthropic compatible endpoint (w/ caching to save $)
… it’s nuts.
Feels like sonnet 4, but literally ~4-8X faster.
Gonna try to combine it with oAuth w/ Max sub to use Opus for planning, Kimi for implementation.
Will see how today plays out.
@AnthropicAI This marks the end of an era. The public may never have such open access to frontier models again.
What happens when only the government and frontier labs have access to the strongest models? I’m not sure, but it’s probably going to be less fun than we’re used to.
@trq212 please release a $1000 plan w/ 100X usage that lets us keep Fable + claude -p from usage credit privileges past the 22nd and I'm immediately buying 2 accs
currently running 3X max subs and am about to run out of limits on all three and I've been trying to limit usage 😭
claude code's 'chapters' feature is one of my favorite QOL changes from the past 6 months.
I run a lot looped workflows where I can step away/go to sleep and come back to the agent still cooking hours later, and chapters have boosted observability there like nothing else.
@icanvardar i've been running and nearly maxxing out 3 codex pro subs weekly since 5.5 dropped, and only had one active claude max acc during that period..
but 4.8's completely won me back. alr back up to two claude max subs, and about to get a third.
i love 5.5, but 4.8 outclasses atm
@codyplof I use a proxy for hermes w/ one of my claude max accounts. Running Opus 4.7 pretty much 24/7, and let it dispatch to kanban tasks w/ smaller models.