Overview
GLM-5.3 is a flagship text model designed for complex software engineering and long-horizon agent tasks. It uses the same base model as GLM-5.2; its improvements come from post-training. It was announced on August 14, 2026. Official announcement. (z.ai)
Capabilities and uses
- Supports coding, terminal operations, multi-step agent workflows and cybersecurity analysis. Zhipu reports a 50% improvement over GLM-5.2 on its internal Z.ai Code Bench and a score of 84.5% on CyberGym; these are publisher-reported benchmark results, not guarantees of real-world performance. (z.ai)
- Supports function calling, streaming, context caching and structured output. The API documentation specifies a 1M-token context window and up to 128K output tokens. (docs.z.ai)
- Model weights are available from Zhipu's official model repository. (huggingface.co)
Limitations and access
- The documented model accepts text-only input; it should not be confused with the multimodal GLM-5.3-Flash. (docs.z.ai)
- Reasoning is always enabled. The
reasoning_effortlevels arelow,highandmax, withmaxas the default. Requests usingthinking.type: "disabled"must be migrated or they will fail. (docs.z.ai) - The weights use the GLM-5.3 license, not the Flash model's MIT license. The license includes a security-review requirement for certain large model-as-a-service businesses; consult the official license for its terms. (huggingface.co)
- Generated code and security findings require independent review and testing.
Official API pricing
Prices below are for GLM-5.3, per 1 million tokens, checked on November 16, 2026. Domestic and international offerings are separate; the figures are not currency conversions.
| Official platform | Currency | Input | Cached input | Output | Cache storage |
|---|---|---|---|---|---|
| Mainland China BigModel API | CNY | 8 | 2 | 28 | Free for a limited time; the table's normal unit is CNY per million tokens per hour |
| International Z.ai API | USD | 1.40 | 0.26 | 4.40 | Free for a limited time |
The domestic table lists a 1M context window. Cached-input rates apply to cache hits; ordinary input, cached input and output are billed separately. Coding Plan subscription points and off-peak discounts are separate from these per-token API prices. Sources: BigModel API pricing and Z.ai API pricing. (docs.bigmodel.cn)
Additional official documentation
International model documentation · Mainland China model documentation