GLM-5.3-Flash

Z.ai's GLM-5.3-Flash was publicly revealed on August 26, 2026, confirmed by Bloomberg, as the anonymous "Ox Alpha" model that had topped OpenRouter and OpenCode usage charts for roughly a week with a 1M-token context window and text/image/video inputs. The model is a 320B-parameter mixture-of-experts with 18B active pe Weights landed on Hugging Face under an MIT license the same evening the identity was confirmed, and Zhipu's Hong Kong shares closed more than 12% higher the next day at roughly 10x January's IPO price. Stripe CEO Patrick Collison had called the still-anonymous model "very impressive" the day before the reveal, with St

South China Morning Post reports that Zhipu AI launched GLM-5.3-Flash, previously code-named Ox Alpha, and that the system ran entirely on a cluster of 100,000 domestically produced Chinese chips during its high-profile stealth trial. The launch positions the model as evidence that China can scale AI on domestic hardwa The coverage frames GLM-5.3-Flash within the broader China-US AI competition narrative, highlighting the model's role in demonstrating domestic-chip capability at scale. The piece connects the technical release to policy goals around chip self-sufficiency and records GLM-5.3-Flash's emergence from anonymity as a signif

Z.ai released GLM-5.3-Flash on August 26, 2026, as the newest member of the GLM-5 family and the first natively multimodal model in the GLM-5 series. The article describes it as a 320 billion total parameter mixture-of-experts model with about 18 billion active parameters per token, supporting text, image, and video un Before the public launch, the model had been previewed anonymously as "Ox Alpha" on OpenRouter and OpenCode, and Z.ai confirmed Ox Alpha was the same model after developers fingerprinted it. The piece positions GLM-5.3-Flash around frontier-level capability at far lower inference cost, noting that the real competition

Ampere's side-by-side comparison shows GLM-5.3 Flash scoring 57 on the Artificial Analysis Intelligence Index against GLM-5.3's 60, while costing around one-ninth of the flagship at normal API pricing and activating only 18B parameters versus 40B. Output speed is measured at roughly 49 tokens per second for Flash again The comparison highlights Flash's distinct advantages: native image input, native multimodal capability, image understanding, visual coding, browser agents, computer-use agents, and 3x usable quota on coding plans. Pricing comes in at $0.15 per million input tokens and $0.50 per million output tokens, versus $1.40 and

Z.ai released GLM-5.3-Flash on August 26, 2026, describing it as the first natively multimodal model in the GLM-5 series and a mixture-of-experts architecture with 320B total parameters and 18B active per token. The model supports a 1,048,576-token context window and accepts image and video input, and weights were publ The model spent its first week running anonymously as "Ox Alpha" on OpenCode and OpenRouter, served entirely on domestically produced Chinese AI chips. Self-hosting guidance in the piece notes the default FP8 checkpoint is roughly 306 GiB before KV cache and that the current vLLM path targets Hopper and newer NVIDIA ha

Local AI Zone's deep dive documents that Z.ai shipped GLM-5.3-Flash on August 26, 2026, as a 320B-parameter mixture-of-experts model activating 18B per token with native FP8 support and a 1,048,576-token context window. It is the first natively multimodal model in the GLM-5 series, and Z.ai's model card explicitly stat The article notes that for twelve days before launch the model had been running anonymously on OpenRouter as "Ox Alpha," a mystery model topping usage charts and prompting reactions from figures like Patrick Collison. GLM-5.3-Flash reportedly beats GLM-5.2 across six coding and agentic benchmarks at roughly one-tenth t

Fello AI's coverage confirms Z.ai released GLM 5.3 Flash on August 26, 2026, with MIT-licensed weights on Hugging Face on day one. The model has 320B total parameters with 18B active, is built on a newly trained base rather than a post-train of GLM 5.2's, and is the first natively multimodal model in the GLM-5 series w The piece reveals that GLM 5.3 Flash is the identity behind Ox Alpha, the anonymous model that had been serving traffic on OpenRouter and drawing attention for its 1M-token context. The article highlights what Z.ai's benchmark chart shows once losing tests are included, and notes that GLM 5.3's promised 744B weights re

DataCamp's feature on GLM-5.3-Flash describes it as Z.ai's cost-optimized model in the GLM-5.3 family, codenamed "Ox Alpha" before its public reveal. The model uses a 320B-A18B MoE design, supports a 1M token context window, and is natively multimodal. It scores 84.3 on Terminal-Bench 2.1, putting it within striking di Artificial Analysis rates GLM-5.3-Flash at an intelligence index of 57 compared to 60 for the full GLM-5.3 flagship. The trade-off is cost and speed: blended API price is roughly $0.10 per 1M tokens versus about $0.90 for GLM-5.3, output speed is about 49 tokens per second, and the 306 GiB FP8 checkpoint rules out ligh