transformers v5.16.1 ships GLM-5.3-Flash, a 320B/18B-active natively multimodal model at one-tenth of GLM-5.2's price

GitHub·high signalXBlueskyLinkedInCopy link

Hugging Face cut v5.16.1 on 2026-08-26 as a special release just to land GLM-5.3-Flash, the first natively multimodal model in the GLM-5 series. It has 320B total parameters with 18B active, trained from a new base model, and introduces a hybrid architecture combining sparse and linear attention to cut long-context serving cost. Z.ai claims it beats GLM-5.2 across benchmarks at one-tenth the price and approaches Claude Opus 4.8 on coding and agentic benchmarks; block/goose v1.48.0 has already swapped GLM-5.2 for GLM-5.3 in its Z.ai provider.