GLM-5.3-Flash is the first natively multimodal model in Z.ai's GLM-5 series: a 320B-parameter Mixture-of-Experts LLM with 18B active parameters per token and a 1M-token context window, built for coding and agentic work.
GLM-5.3-Flash pairs sparse attention with linear attention in a hybrid architecture, a first for the GLM series, cutting attention computation 3× and KV cache size 4.4× compared with GLM-5.3. Manifold-Constrained Hyper-Connections (mHC) further improve scaling efficiency. Trained on a 30T-token multimodal corpus, it outperforms GLM-5.2 on coding and agentic benchmarks at one-tenth the price, and approaches Claude Opus 4.8 overall, with scores of 84.3 on Terminal-Bench 2.1 and 63.4 on DeepSWE v1.1. Full details are in Z.ai's announcement.
Visual capabilities are natively integrated into the coding loop: the model can observe interfaces, rendered results, and interaction feedback, then test and refine its own work, from frontend development and game creation to Blender 3D scenes and computer-use tasks.
On a Shared Endpoint, you pay per token. The endpoint is OpenAI-compatible and already live: point your existing SDK at it and start sending requests.