GLM-5.3-Flash

Linas Beliūnas's newsletter reports that an anonymous listing called Ox Alpha appeared on OpenRouter and OpenCode on August 20, 2026 with a 1-million-token context window, text, image, and video input, and no listed owner. Over six days it became the most-used model on OpenRouter, processing roughly 23 trillion tokens, The same piece describes GLM-5.3-Flash as the first natively multimodal model in the GLM-5 series and notes that Zhipu's Hong Kong shares closed more than 12 percent higher the day after launch, roughly ten times the January IPO price. It frames the model as a reset for cost-performance economics because Artificial Ana

Fello AI's coverage confirms that Z.ai shipped GLM-5.3-Flash on August 26, 2026, with MIT-licensed weights on Hugging Face on day one, as a 320B-total, 18B-active mixture-of-experts built on a newly trained base rather than a post-train of GLM-5.2. The article frames GLM-5.3-Flash as the first natively multimodal model The same piece surfaces Z.ai's launch pricing of $0.15 per million input tokens and $0.50 per million output for GLM-5.3-Flash, versus $1.40 and $4.40 for the flagship GLM-5.3, and discusses the Ox Alpha identity reveal, in which the anonymous model that had been serving traffic on OpenRouter for about twelve days was

AI Profit Boardroom's guide reports that Z.ai released GLM-5.3-Flash on August 26, 2026 as a 320-billion-parameter mixture-of-experts model with 18 billion active parameters, native multimodal input for text, images, and video, and a one-million-token context window under the MIT license. It notes that the listed OpenR The same piece recounts the Ox Alpha preview period in which the anonymous model offered near-free access on OpenRouter from August 20, 2026 onward, with the community fingerprinting it back to the GLM family within 48 hours before Z.ai's August 26 announcement confirmed ownership. It walks through two genuine ways to

A Medium piece by Greek Ai on CodeToDeploy reports that Z.ai released GLM-5.3-Flash on August 26, 2026 as the newest member of the GLM-5 family, designed for frontier-level capabilities without frontier-level inference costs. It describes the model as having 320 billion total parameters with about 18 billion active per The same article frames the launch around the Ox Alpha reveal: before Z.ai officially announced GLM-5.3-Flash, developers were experimenting with a mysterious model of that name on platforms such as OpenRouter and OpenCode to collect real-world feedback, and Z.ai later confirmed Ox Alpha was an anonymous preview of GLM

Ampere.sh's head-to-head comparison reports that Z.ai's GLM-5.3-Flash is a 320B-total, 18B-active mixture-of-experts model with a 1-million-token context window, native image input, and a $0.15/$0.50 per million token API price, priced at roughly one-ninth the flagship GLM-5.3's $1.40/$4.40 per million tokens. Artifici The comparison also notes that GLM-5.3-Flash is not faster at token generation: Artificial Analysis measures GLM-5.3 at about 85 tokens per second versus about 49 tokens per second for Flash, with a slightly lower time to first token of 1.46 seconds. It concludes that GLM-5.3-Flash is the better value for most high-vol

A community pull request on the official zai-org/GLM-5.3-Flash Hugging Face repository, opened on August 26, 2026 and merged shortly after, adds evaluation results extracted from the model card's benchmark chart in HF Hub-compatible YAML form. The merged results list Terminal-Bench 2.1 at 84.3, DeepSWE at 63.4, and Hum The discussion also flags a sourcing caveat: the model card cites arxiv:2602.15763 as 'the GLM-5 Technical report,' but that paper was published in February 2026 and predates GLM-5.3-Flash, so it does not contain variant-specific numbers and was not used as a source for the new evaluation entries. The thread confirms t

Z.ai released GLM-5.3-Flash on August 26, 2026, as a 320-billion-parameter mixture-of-experts model that activates only 18 billion parameters per token and runs natively in FP8. According to a technical deep dive by Local AI Zone citing the Z.ai model card, it is the first natively multimodal model in the GLM-5 series, The same deep dive reports that GLM-5.3-Flash beats GLM-5.2 across six coding and agentic benchmarks at roughly one-tenth the price and approaches Claude Opus 4.8 on long-horizon agent tasks. For twelve days before launch, an anonymous model called Ox Alpha was the most-used model on OpenRouter and topped OpenCode usag

Artificial Analysis published an independent evaluation page for GLM-5.3-Flash, scoring it 46 on the Artificial Analysis Intelligence Index, placing it well above the median of 22 among 112 comparable open-weight models. The model supports text and image input, produces text output, has a 1M-token context window, is bu On cost and speed, the page reports $0.15 per million input tokens and $0.50 per million output tokens — described as moderately priced — with an 83% cache discount bringing effective output cost to roughly $0.18. Output throughput is approximately 48.5 tokens per second, which Artificial Analysis flags as notably slow