Hacker News Reader

I’m kind of lucky that most of the time I don’t have to use it through peak times so it’s not that bad speed wise

The legacy plan I have is so good as to be basically unlimited usage for my workloads, so I’m kind of stuck with it til they stop renewing it haha

Cache cost is the primary metric, imo. Most people look at output cost, but you only pay that once. Cache cost you pay every turn and it grows each turn as the context length grows.

The biggest problem with DeepSeek 4.1 Flash is all that it hallucinates a lot.

[1] - https://www.youtube.com/watch?v=P4dTq4X8bqk