GLM-5.3-Flash matches top models at 1/7th cost, runs without Nvidia
The Decoder · rss · 2026-08-27
Z.ai releases GLM-5.3-Flash, an open-source model with 320 billion parameters. It lands just three points behind the larger GLM-5.3 on Artificial Analysis's Intelligence Index while costing a seventh as much. Notably, all inference traffic ran on Chinese AI chips instead of Nvidia hardware, demonstrating high performance and low cost in a non-Nvidia ecosystem.
More from Infra
- Kioxia to build new chip plant in northern Japan as AI storage demand climbs — ssh4net · 2026-08-27
- Trick: Bypass Google's Encrypted /goto Links via Language Header — gaganghotra_ · 2026-08-27
- Nvidia Rarely Guides for 70% Growth in Fiscal 2028 Revenue — heypearlai · 2026-08-27
- Proposal: Using BitTorrent to Host Large AI Model Files — vexatious-big · 2026-08-27
- Anthropic locks in $45B compute deal with Nscale ahead of IPO — The Decoder · 2026-08-27
- Hugging Face staff clarifies role: supporting open source & enterprise hosting — mervenoyann · 2026-08-27