Zai releases GLM-5.3 Flash: 320B params, 2x efficiency on DeepSWE

togethercompute · x · 2026-08-28

Zai has released GLM-5.3 Flash, a natively multimodal model with 320B total parameters, 18B active, 1M context, and hybrid attention. On the DeepSWE benchmark, it nearly matches Luna's performance while completing more than twice the work for the same budget.

Related event: Zai Open-Sources GLM-5.3 Flash, a 320B Native Multimodal Model(3 posts)→

Original post →

More from Models

Models channel →