GLM-5.3 is a large-scale reasoning model from Z.ai, built for complex software engineering and long-horizon agent tasks. It supports text input and output with a 1M-token context window, and improves on GLM-5.2 in coding and in the balance between performance and token efficiency.
Reasoning is always on and cannot be disabled. Reasoning efforts low, high, and max are supported; max is the default.
提供商
不同公司托管同一模型。OpenRouter 根据您选择的路由模式将请求路由到其中一家公司:Balanced(价格 + 速度)、Nitro(速度最快)或 Floor(价格最低)。
| US$0.281 | US$1.25 | US$0.062 | -- | -- | ||
| US$0.72US$0.450 | US$3.20US$2.00 | US$0.16US$0.100 | -- | -- | ||
| US$0.700 | US$2.20 | US$0.130 | -- | -- |
- 输入 US$0.281 / 输出 US$1.25-- tok/s-- 正常运行时间
输入 US$0.45 / 输出 US$2.0038% off-- tok/s-- 正常运行时间- 输入 US$0.70 / 输出 US$2.20-- tok/s-- 正常运行时间
定价
客户为此模型实际支付的平均价格,与提供商公布的价格并列展示。由于缓存和折扣,实际支付的价格往往远低于标价。
GLM 5.3 (batch) — 价格历史
| US$0.281 | US$1.25 | US$0.062 | — | ||
| US$0.450 | US$2.00 | US$0.100 | — | ||
| US$0.700 | US$2.20 | US$0.130 | — |
- 输入 US$0.281 / 输出 US$1.25
- 输入 US$0.450 / 输出 US$2.00
- 输入 US$0.700 / 输出 US$2.20
性能
吞吐量是模型生成内容的速度(每秒 token 数,越高越好)。延迟是完整的往返时间(越低越好)。TTFT 是首个 token 的响应时间,即您看到内容出现前需要等待的时间(越低越好)。
正常运行时间
正常运行时间是指过去 3 天中至少有一家提供商响应请求的时间百分比。可用性是指成功提供推理服务的时间百分比。OpenRouter 会持续监控,并在某家提供商返回错误时使用次优提供商。
正常运行时间数据尚不足,暂时无法显示。
当上游提供商发生错误时,如果您的请求筛选条件允许,我们可以通过路由到另一家健康的提供商来恢复服务。您可以通过以下接口以编程方式访问各提供商的正常运行时间数据: 端点 API。了解更多有关我们的 负载均衡和提供商路由选项。
基准测试
标准化评测得分,以及此模型针对共用基准测试提示词的实际输出。百分比越高越好,每项排名显示此模型在该来源所评测的模型中的位置。
| 来源 | 基准测试 | 得分 |
|---|---|---|
| Artificial Analysis | GLM-5.3 (Low) Intelligence Index | 23.1 |
| Artificial Analysis | GLM-5.3 (Low) Finance & Accounting Index | 31.2 |
| Artificial Analysis | GLM-5.3 (Low) Strategy & Ops Index | 38.0 |
| Artificial Analysis | GLM-5.3 (Low) Legal Index | 27.7 |
| Artificial Analysis | GLM-5.3 (Low) Healthcare & Medical Index | 31.0 |
| Artificial Analysis | GLM-5.3 (Low) Engineering Index | 28.3 |
| Artificial Analysis | GLM-5.3 (Low) Economics Index | 28.6 |
| Artificial Analysis | GLM-5.3 (Low) Openness Index | 33.3 |
| Artificial Analysis | GLM-5.3 (Low) AutomationBench-AA | 52.0% |
| Artificial Analysis | GLM-5.3 (Low) GDP.pdf | 8.8% |
| Artificial Analysis | GLM-5.3 (Low) HLE | 36.6% |
| Artificial Analysis | GLM-5.3 (Low) AA-LCR | 72.7% |
| Artificial Analysis | GLM-5.3 (Low) GDPval-AA | 41.6% |
| Artificial Analysis | GLM-5.3 (Low) CritPt | 14.6% |
| Artificial Analysis | GLM-5.3 (Low) HLE-Diamond | 17.5% |
| Artificial Analysis | GLM-5.3 (Low) SciCode | 42.0% |
| Artificial Analysis | GLM-5.3 (Low) Terminal-Bench 4.0 | 34.8% |
| Artificial Analysis | GLM-5.3 (Low) FrontierCode-AA 2.0 | 12.7% |
| Artificial Analysis | GLM-5.3 (Low) FrontierCode 1.1 | 34.9% |
| Artificial Analysis | GLM-5.3 (Low) Terminal-Bench-Science | 3.8% |
| Artificial Analysis | GLM-5.3 (Low) CritPt v2 (Hard Subset) | 21.5% |
| Artificial Analysis | GLM-5.3 (Low) CritPt v2 | 45.7% |
| Artificial Analysis | GLM-5.3 (Low) AA-Omniscience Accuracy | 31.9% |
| Artificial Analysis | GLM-5.3 (Low) AA-Omniscience Non-Hallucination Rate | 36.8% |
| Artificial Analysis | GLM-5.3 (Low) MLCR-AA | 11.1% |
| Artificial Analysis | GLM-5.3 (Max) Intelligence Index | 31.7 |
| Artificial Analysis | GLM-5.3 (Max) Finance & Accounting Index | 42.1 |
| Artificial Analysis | GLM-5.3 (Max) Strategy & Ops Index | 47.4 |
| Artificial Analysis | GLM-5.3 (Max) Legal Index | 35.8 |
| Artificial Analysis | GLM-5.3 (Max) Healthcare & Medical Index | 39.3 |
| Artificial Analysis | GLM-5.3 (Max) Engineering Index | 37.6 |
| Artificial Analysis | GLM-5.3 (Max) Economics Index | 37.3 |
| Artificial Analysis | GLM-5.3 (Max) Cyber Index | 36.4 |
| Artificial Analysis | GLM-5.3 (Max) Openness Index | 33.3 |
| Artificial Analysis | GLM-5.3 (Max) AutomationBench-AA | 62.2% |
| Artificial Analysis | GLM-5.3 (Max) GDP.pdf | 11.2% |
| Artificial Analysis | GLM-5.3 (Max) AA-AnalystAgent | 38.8% |
| Artificial Analysis | GLM-5.3 (Max) EnterpriseOps-Gym-AA | 36.4% |
| Artificial Analysis | GLM-5.3 (Max) ITBench-AA | 46.1% |
| Artificial Analysis | GLM-5.3 (Max) Harvey LAB-AA | 0.3% |
| Artificial Analysis | GLM-5.3 (Max) GPQA Diamond | 91.7% |
| Artificial Analysis | GLM-5.3 (Max) HLE | 42.3% |
| Artificial Analysis | GLM-5.3 (Max) AA-LCR | 79.7% |
| Artificial Analysis | GLM-5.3 (Max) τ³-Banking | 50.3% |
| Artificial Analysis | GLM-5.3 (Max) GDPval-AA | 57.7% |
| Artificial Analysis | GLM-5.3 (Max) CritPt | 19.1% |
| Artificial Analysis | GLM-5.3 (Max) HLE-Diamond | 17.9% |
| Artificial Analysis | GLM-5.3 (Max) SciCode | 59.0% |
| Artificial Analysis | GLM-5.3 (Max) Terminal-Bench 4.0 | 41.9% |
| Artificial Analysis | GLM-5.3 (Max) Terminal-Bench 2.1 | 83.9% |
| Artificial Analysis | GLM-5.3 (Max) FrontierCode-AA 2.0 | 21.6% |
| Artificial Analysis | GLM-5.3 (Max) FrontierCode 1.1 | 40.1% |
| Artificial Analysis | GLM-5.3 (Max) Terminal-Bench-Science | 9.5% |
| Artificial Analysis | GLM-5.3 (Max) CritPt v2 (Hard Subset) | 33.5% |
| Artificial Analysis | GLM-5.3 (Max) CritPt v2 | 57.4% |
| Artificial Analysis | GLM-5.3 (Max) AA-Omniscience Accuracy | 32.3% |
| Artificial Analysis | GLM-5.3 (Max) AA-Omniscience Non-Hallucination Rate | 61.1% |
| Artificial Analysis | GLM-5.3 (Max) CWE-Bench-AA | 55.8% |
| Artificial Analysis | GLM-5.3 (Max) DeepsecBench-AA | 24.3% |
| Artificial Analysis | GLM-5.3 (Max) CyberGym-E2E-AA | 29.0% |
| Artificial Analysis | GLM-5.3 (Max) MLCR-AA | 17.2% |
| Artificial Analysis | GLM-5.3 (Max) HealthBench | 54.5% |
| Artificial Analysis | GLM-5.3 (Max) HealthBench Professional | 52.8% |
| Design Arena | GLM-5.3 Models Arena 3D Elo | 1285 |
| Design Arena | GLM-5.3 Models Arena Asciiart Elo | 1180 |
| Design Arena | GLM-5.3 Models Arena Code Categories Elo | 1274 |
| Design Arena | GLM-5.3 Models Arena Data Visualization Elo | 1248 |
| Design Arena | GLM-5.3 Models Arena Game Development Elo | 1279 |
| Design Arena | GLM-5.3 Models Arena SVG Elo | 1239 |
| Design Arena | GLM-5.3 Models Arena UI Component Elo | 1282 |
| Design Arena | GLM-5.3 Models Arena Website Elo | 1275 |
| Design Arena | GLM-5.3 Agents Arena Htmlslides Elo | 1189 |
| Design Arena | GLM-5.3 Agents Arena Mobile Apps Elo | 1184 |
| Design Arena | GLM-5.3 Agents Arena Python-Pptxslides Elo | 1217 |
| Design Arena | GLM-5.3 Agents Arena Webapps Elo | 1270 |
| Vals AI | GLM 5.3 Access to Justice Benchmark | 73.7% |
| Vals AI | GLM 5.3 Code Migration v1.1 | 55.1% |
| Vals AI | GLM 5.3 CyberBench v1.1 | 72.1% |
| Vals AI | GLM 5.3 EMB v1.1 | 69.1% |
| Vals AI | GLM 5.3 Finance Agent v2.1 | 65.0% |
| Vals AI | GLM 5.3 Vals Finance Index | 61.4% |
| Vals AI | GLM 5.3 Financial Coaching Bench | 74.1% |
| Vals AI | GLM 5.3 GPQA Diamond | 88.1% |
| Vals AI | GLM 5.3 HLAB-Vals-1.1 | 14.6% |
| Vals AI | GLM 5.3 Immigration | 66.7% |
| Vals AI | GLM 5.3 IOI | 68.4% |
| Vals AI | GLM 5.3 LiveCodeBench | 80.5% |
| Vals AI | GLM 5.3 LegalBench | 84.8% |
| Vals AI | GLM 5.3 Vals Legal Index | 74.6% |
| Vals AI | GLM 5.3 Legal Research Bench | 49.0% |
| Vals AI | GLM 5.3 MedCode | 42.9% |
| Vals AI | GLM 5.3 MedScribe | 88.8% |
| Vals AI | GLM 5.3 MMLU Pro | 86.8% |
| Vals AI | GLM 5.3 MysteryMechanism | 23.0% |
| Vals AI | GLM 5.3 ProgramBench | 1.5% |
| Vals AI | GLM 5.3 ProofBench v1.1 | 49.0% |
| Vals AI | GLM 5.3 Public Benefits Bench v1.1 | 68.5% |
| Vals AI | GLM 5.3 Vals RSI Index | 22.9% |
| Vals AI | GLM 5.3 SkillsBench | 47.5% |
| Vals AI | GLM 5.3 SRE Bench | 9.2% |
| Vals AI | GLM 5.3 SWE-bench | 95.4% |
| Vals AI | GLM 5.3 Tax Agent Bench | 43.3% |
| Vals AI | GLM 5.3 TaxEval v2 | 72.4% |
| Vals AI | GLM 5.3 Terminal-Bench 4.0 | 38.9% |
| Vals AI | GLM 5.3 Terminal-Bench-Science | 5.7% |
| Vals AI | GLM 5.3 Vals Index | 59.8% |
| Vals AI | GLM 5.3 Vibe Code Bench 1-100 v1.1 | 29.8% |
| Vals AI | GLM 5.3 Vibe Code Bench v1.1 | 78.1% |
应用
向此模型发送最多流量的公开应用。这能很好地反映实际生产工作负载的情况,也能提示此模型最适合哪些用例。
使用此模型的应用尚无足够数据。
活动
此模型的 token 用量和请求流量随时间的变化。
API
可直接用于调用此模型的代码。OpenRouter 的 API 与 OpenAI 兼容——大多数 SDK 只需替换基础 URL 即可使用。切换模型时,唯一需要更改的是下方的模型标识符。
目前无法通过 EU 区域内路由使用。 区域内路由文档
文档
在 OpenRouter 上使用 Z.ai: GLM 5.3 (batch) 构建应用的指南、实用教程、博客文章和工具。
常见问题
GLM-5.3 is a large-scale reasoning model from Z.ai, built for complex software engineering and long-horizon agent tasks. It supports text input and output with a 1M-token context window, and improves on GLM-5.2 in coding and in the balance between performance and token efficiency. Reasoning is always on and cannot be disabled. Reasoning efforts low, high, and max are supported; max is the default.
OpenRouter 上的 GLM 5.3 (batch) 文档包括 Self-Hosted LLM vs API: When Does Self-Hosting Pay Off 博客文章和How to Benchmark LLMs 博客文章。快速入门介绍了 API 密钥和您的首次请求,API 参考列出了所有参数。
GLM 5.3 (batch) 的费用为 US$0.281/百万输入 token 和 US$1.25/百万输出 token,Cache Read,价格为 US$0.062/M tokens 单独计费。
GLM 5.3 (batch) 的上下文窗口为 1,048,576 个 token。它最多支持 1,048,576 个补全 token。
是的。GLM 5.3 (batch) 接受 tools 和 tool_choice,以进行函数调用。 它还支持通过 response_format 中的 JSON schema 实现结构化输出。
在 OpenRouter 上,GLM 5.3 (batch) 由 3 家提供商提供服务:Morph、DeepInfra和Together。 请求会路由至最佳的可用提供商,并在故障时自动切换至其他提供商,您也可以通过提供商路由固定或排除提供商。
GLM 5.3 (batch) 于 2026年8月18日 发布。