GLM-5.3-Flash

GLM-5.3-Flash is the first natively multimodal model of the GLM-5 family, published by Z.ai on 26 August 2026. The name says cheap variant, but in this family the roles are reversed: this is the model whose weights anyone can download, while the flagship GLM-5.3 from 14 August is sold only through the API and the coding subscription. Z.ai describes 320 billion total parameters with 18 billion active per token; the safetensors index on Hugging Face totals 321.3 billion, so the declaration checks out. For scale, the last flagship whose weights were released, GLM-5.2 from June 2026, has 753.3 billion — this model is less than half its size. The architecture is the interesting part. Z.ai calls it the first open-weight frontier model to combine sparse attention with linear attention, and gives the reason in numbers: against GLM-5.3 it cuts attention computation 3.01 times and the KV cache 4.44 times. The company adds Manifold-Constrained Hyper-Connections and a 30-trillion-token multimodal pre-training corpus. Vision is built into the coding loop rather than bolted on: the model looks at interfaces and rendered output, then corrects its own work. The price follows the architecture. The list rate is 0.15 dollars per million input tokens and 0.50 per million output, against 1.40 and 4.40 for GLM-5.3 — 9.3 times cheaper on input and 8.8 times cheaper on output. Until 9 September 2026 a 50 percent promotion halves that again. Z.ai claims the model beats GLM-5.2 across benchmarks at one tenth of the price and approaches Claude Opus 4.8 on coding and agentic tests; the company publishes those results only as a chart, and no independent laboratory has repeated them. The licence is worth stating plainly, because model cards often are not what their label says: the LICENSE file in the repository is the bare MIT text, with no attribution rider and no field-of-use limits.

▸News

Z.ai opened the weights of its cheap model — the flagship it promised two weeks ago is still closed

8/28/2026 · Z.AI Developer Documentation / Hugging Face

▸Videos

No videos yet.

▸Reviews