Zai open-sources 320B GLM-5.3, claims full training on domestic chips
SumitGup · x · 2026-08-27
Zai has released GLM-5.3-Flash (previously Ox Alpha), a 320B-parameter MoE model (18B active) featuring native multimodality and a 1M-token context window under the MIT license. The company emphasizes that the model was trained and is served entirely on domestic Chinese AI chips, demonstrating significant infrastructure capabilities amidst hardware constraints. Weights, APIs, and implementation examples for coding and tool use are now available across official platforms.
Related event: Zhipu Open-Sources GLM-5.3-Flash with Frontier Performance at Low Cost(48 posts)→
More from Infra
- Zhipu's GLM-5.3-Flash Served Record Traffic Entirely on Domestic Chinese Chips — teortaxesTex · 2026-08-27
- Benedict Evans: Nvidia's cash flow is reshaping the AI chip ecosystem — SumitGup · 2026-08-27
- UK grid jammed by phantom data centers; Ofgem plans deposits up to hundreds of millions — nordicinst · 2026-08-27
- OCP Report: Copper at $0.05/Gbps vs Optics at $0.5/Gbps — jwt0625 · 2026-08-27
- Tokyo AI Event Explores Agentic Memory and Retrieval on Edge Devices — Stefania_druga · 2026-08-27
- Huawei Starts Exporting Ascend Chips, Extending Its Sovereign AI Play Beyond China — kevinsxu · 2026-08-27