Hassan
@buildwithhassan
SWE, AI Automation Enthusiast. I test every AI coding agent so you don’t have to, what’s worth using, what it really costs, and what the docs won’t tell you.
336 Following 1.8K Followers
I've been trying to figure out how much usage Kimi's top plan actually gives you, and the numbers I found make it look even worse.
At a 97% cache hit rate, the roughly $99 plan works out to about 660 million tokens a week on K3.1. That's actually up from under 500 million on K3, so the price did get better.
But a $20 Claude plan gets close to 800 million tokens a week, and that's on Opus 5.5.
So you're paying almost five times more for less usage on a model that still ranks below Opus. That's why I keep saying K3.1 is a great model with the wrong subscription.
Show more
OpenCode's sitemap just leaked GLM-5.5 Flash, GLM-5.4, Kimi K4, DeepSeek V4.1 Pro and Muse Spark 1.4 before anyone announced them.
They removed them from the sitemap rn, but the pages are still live. Kimi K4 already has its own stats page sitting at 0 tokens, waiting for launch day.
The best AI leak source in 2026 is apparently an SEO file.
Show more
notice this one detail on Z AI's pricing page
both GLM-5.3 and 5.3 Flash show the cache storage as "limited-time free"
so cache storage is free for now, but Z AI is clearly leaving the door open to charging for it later.
kimi K3 on bedrock already charges $3.75 for cache writes. if Z AI follows that direction, the effective cost of GLM jumps overnight without the listed price changing
when you're evaluating model pricing, the cache price is the one that can change the economics later
Show more
