A Free Model Just Tied Claude Opus 4.8

A free, open, MIT-licensed model just reportedly tied Claude Opus 4.8 on the hardest long-horizon coding benchmark — at about one sixth the cost. This is GLM-5.2 from Z.ai (formerly Zhipu): a 744B-param Mixture-of-Experts coder with a 1,000,000-token context, open weights on Hugging Face + ModelScope, and a catch most coverage is skipping — the hosted API is China-hosted, so the privacy answer is to run the open weights yourself. Benchmarks are reported/published figures; independent verification is still early.

Sources: VentureBeat, TechTimes, tonyreviewsthings.com, cryptobriefing.com, ChatForest.

Видео A Free Model Just Tied Claude Opus 4.8 канала BuildBeat