Live drop
Z.aiDropped Aug 26, 20264 builds4 makers2 prompts
Natively multimodal 320B-A18B MoE with hybrid sparse/linear attention for low-cost agentic coding and vision tasks.
About this release
GLM-5.3-Flash is Z.ai's first natively multimodal model in the GLM-5 series, released under MIT license with open weights. It uses a hybrid sparse + linear attention architecture (320B total, 18B active) plus Manifold-Constrained Hyper-Connections on a 30T multimodal pretraining corpus, achieving 57 on the Artificial Analysis Intelligence Index while approaching Claude Opus 4.8 on coding and agentic benchmarks at a fraction of the cost of GLM-5.3, with 1M context and native image/video support.
Read the official Z.ai launch pageWorth opening first
Same brief, other models
All challenges →Challenge
Same brief across models — older takes on the left, newer on the right.
Paste the URL and Dropday.ai will credit you, track the build as it evolves, and put it on this drop.
Submit a build