Post

Ed
@ed3d.net

AA's numbers are generally pretty well matched to the relative vibes, except that they really overrate frontier midrange models (Terra, Sonnet) They do not include tests for long-haul directional adherence, though, which is a major gap You will not notice in practice a 5-point swing between models

Techmeme@techmeme.com

Z.ai's GLM-5.3 with max reasoning scores 60 on the Artificial Analysis Intelligence Index, on par with Kimi K3 but below Opus 5 at 63 and Fable 5 at 62 (@artificialanlys) Main Link | Techmeme Permalink

Ed
@ed3d.net

I think GLM-5.3 is fine. I also think it might be overcooked. It talks like Opus 5 and that is a bad thing. I also have had more off-task sessions with it as my top orchestrator than 5.2. My working assumption is that as a reviewer and a planning oracle it is very substitutable for Sol.

Ed
@ed3d.net

On Max, GLM-5.3's token usage is best described as "bullshit" and the high/max performance divide seems very small. Polytoken defaults both 5.2 and 5.3 to high, because I am ungovernable by model provider advice.

1 repost8 likes

No replies yet.