What's this about?
Tencent released and open-sourced Hy4 preview. Within a couple of days it was reportedly processing trillions of tokens on OpenRouter — more than GLM 5.3 managed in a week.
Why would a new model get that much traffic that fast?
Price, mostly. The detail commenters point to is a 5% cache cost when most providers charge 10–20% for cached input tokens. For anything with a long stable prefix — a system prompt, a codebase, a document set — cache pricing dominates the bill, so a model that is merely competitive on quality but far cheaper on cache can pick up enormous volume immediately without being better at anything.
What's the community saying?
HN 380 points, 239 comments
- The traction numbers are the lead, with the cache-pricing explanation attached.
- Immediate scepticism about whether OpenRouter rankings are being gamed by publishers paying for exposure — nobody in the thread assumes otherwise.
- Which turns, as these threads do, into a long argument about whether any of this counts as capitalism; skip that portion.