Digest · both · chinese ai news

This week in Chinese AI: prices checked, two list prices that lie

Published 16 Oct 2026. Updated 16 Oct 2026. Researched 16 Oct 2026.

This is the first weekly note. There is no prior snapshot in the public history to diff, so “what changed” would be a guess. What we did check is below. Later weeks can subtract.

Prices

The build’s OpenRouter pull is the source of every percentage on the site. Flash models Qwen3.8 Flash, GLM-5.3 Flash, and Hunyuan HY3 cluster at the same blended discount. That sentence will be rewritten by the chart if the snapshot moves.

The correction we are making up front

GLM-5.3’s OpenRouter card is not Z.AI’s $1.40 / $4.40. Kimi K3’s card is not Moonshot’s $3 / $15. Both labs’ own endpoints on OpenRouter match the docs. Details: OpenRouter versus direct.

Benchmarks stored

LMArena style-control, publish date on the rows: 15 Oct 2026. SWE-bench Verified JSON fetched 16 Oct 2026. The Verified file still has no K3, no GLM-5.3, and no DeepSeek V4 row in the names we searched. Terminal-Bench numbers on the site are Moonshot’s card, with the harness caveat in the Kimi guide.

Not in this week

No hands-on sign-up. No video-credit grid for Kling or Hailuo. No Artificial Analysis pull, because that API needs a key. No translated WeChat item, because we did not archive a primary Chinese post this week. A digest that invents one would be worse than a short one.

Next checks worth doing: a Qwen Model Studio free-quota screenshot, a Kling credit grid saved into the repo, and a second price snapshot so this section can be a diff.