GLM-5.3-Flash will likely handle 45% of your AI workloads
ThinkingNews Desk · how this was written
Z.ai has released GLM-5.3-Flash, the first natively multimodal model in the GLM-5 series, featuring 320 billion parameters. The company says the model outperforms GLM-5.2 while costing roughly one-tenth as much, and analysts estimate it could handle about 45 percent of typical AI workloads.
Written from all 3 reports below, not from any single one.
How it was reported
- Hacker News·GLM-5.3-Flash
- TechMeme·Z.ai releases GLM-5.3-Flash, the first natively multimodal GLM-5 series model, with 320B parameters, saying it outperforms GLM-5.2 at "one-tenth the price" (Z.ai)
Z.ai unveiled GLM-5.3-Flash, a 320-billion-parameter multimodal model that the company claims surpasses its predecessor GLM-5.2 while costing roughly one-tenth as much to run. In the same period, Nvidia reported Q2 revenue of $96.22 billion, a 106% year-over-year increase, with data-center sales rising 117% to $89 billion.
- VentureBeat·GLM-5.3-Flash will likely handle 45% of your AI workloads
GLM-5.3-Flash, identified as the mystery “Ox Alpha” model, is hosted by Z.ai on Chinese chips and priced at $0.15-$0.50 per million tokens (promo 7.5-25 cents). Its open-weight MIT license and strong performance place it at 57 on Artificial Analysis’s intelligence-vs-cost index, offering roughly one-tenth the cost of comparable U.S. mid-tier models. Enterprises such as Uber are already feeling budget strain, prompting a shift toward cheaper Chinese inference providers.