Zhipu Releases GLM 5.3: 100T Daily Compute on Domestic Chips, Costs Slashed by 90%
oran_ge · x · 2026-08-26
Zhipu released the GLM 5.3 Flash Vision model, featuring a 320B parameter architecture (18B active).
Key Highlights:
- Infrastructure: Global daily supply exceeds 100T compute, fully powered by domestic chips. End-to-end service performance improved by 3x, achieving per-token cost parity with mainstream Nvidia GPUs.
- Performance: Scored 57 on the Artificial Analysis Intelligence Index, tying with Opus 4.8.
- Pricing: Limited-time 50% discount, costing just 1/40 of Opus 4.8 and 1/10 of the previous GLM 5.3 generation.
- Multimodality: Native multimodal capabilities allowing iterative visual improvements, including generating kitchen scenes autonomously in Blender.
The model is now live on the official website.
More from Infra
- Proposal: DGX Spark-class devices sold on $200/mo contracts could mesh into a giant cheap inference network — jasonkneen · 2026-08-27
- Glean reveals model routing scores: GPT-5.6 Luna leads at $0.08 — testingcatalog · 2026-08-27
- Serving frontier models at scale on purely Chinese hardware — tokumin · 2026-08-27
- Firecrawl Launches Startup Deal: Up to $30k in Credits — devdigest · 2026-08-27
- Long Read: AI Is Buying the Data of Dead Companies — rvp · 2026-08-27
- Antirez: High prefill speed makes LLMs feel 10x more powerful — antirez · 2026-08-27