Hacker News Reader

v4 flash was really, really good in practice for me. While on openrouter it's around 1/100th of what the "SOTA" models cost.

But billions? A bit exaggerated.

Plenty of screenshots in this thread showing usage.

Its the ridiculous cached input token price of $0.00028/1 M

So basically forcing particular deepseek provider should make cache introduce another huge savings bump?

Count me in, I'll be testing that. Thank you!