GLM-5.3-Flash 是 智谱AI 推出的对话大模型,发布于 2026-08。GLM-5.3-Flash is a Chat Model released by 智谱AI, launched on 2026-08.
智谱(Z.ai)2026-08-26 揭晓的匿名模型「Ox Alpha」真身:320B 总参 / 18B 激活的 MoE,GLM-5 系列首个原生多模态成员(文本、图像、视频输入),1M 上下文、128K 输出,权重以 MIT 协议开源可商用。标准价 $0.15/$0.50 每百万 tokens(缓存 $0.015),发布促销五折至 2026-09-09($0.075/$0.25)。DeepSWE v1.1 63.4、AutomationBench 48.8,编程与智能体表现逼近 Claude Opus 4.8,匿名预览期的全部推理流量均跑在国产芯片上。Revealed by Z.ai on 2026-08-26 as the true identity of the anonymous 'Ox Alpha': a 320B-total / 18B-active MoE and the first natively multimodal member of the GLM-5 series (text, image and video input), with 1M context and 128K output, released under an MIT license. List price $0.15/$0.50 per million tokens (cache $0.015), halved to $0.075/$0.25 until 2026-09-09. Scores 63.4 on DeepSWE v1.1 and 48.8 on AutomationBench, approaching Claude Opus 4.8 on coding and agentic work; all preview traffic was served on domestically produced chips.
MIT 开源可商用,自部署无限制、1M 上下文 + 原生视频/图像输入、促销价 $0.075 极致便宜MIT license, unrestricted self-hosting、1M context with native video/image input、Promo price as low as $0.075
注意事项
促销价 9/9 后翻倍、跑分多为官方自测、超长上下文成本仍随长度上升Promo pricing doubles after Sep 9、Mostly self-reported benchmarks、Long-context cost still scales with length
关于 GLM-5.3-Flash 的常见问题Frequently Asked Questions about GLM-5.3-Flash
GLM-5.3-Flash 是免费的吗?Is GLM-5.3-Flash free?
GLM-5.3-Flash 目前以付费 API / 平台调用为主,可关注官方是否提供免费试用额度。GLM-5.3-Flash is mainly available via paid API/platform calls; watch for official free trial quotas.
GLM-5.3-Flash 的上下文窗口有多大?How large is GLM-5.3-Flash's context window?
GLM-5.3-Flash 的上下文窗口为 1M,适合处理长文档与多轮复杂对话。GLM-5.3-Flash's context window is 1M, suitable for long documents and multi-turn complex dialogue.
GLM-5.3-Flash 是开源还是闭源?Is GLM-5.3-Flash open-source or closed-source?
开源模型,可通过官方开源仓库或自部署方式使用。Open-source model, available via official repositories or self-hosting.
GLM-5.3-Flash 适合用来做什么?What is GLM-5.3-Flash good for?
GLM-5.3-Flash 的主要能力包括 chat、vision、video、function、reasoning、agent,可用于相关场景的创作、分析与自动化任务。GLM-5.3-Flash's main capabilities include chat、vision、video、function、reasoning、agent, usable for creation, analysis and automation in related scenarios.