GLM-5.3

GLM-5.3 is designed for complex coding and long-horizon work, building on the same foundation as its predecessor while applying additional post-training. The reported improvements come from expanding a training stack that supports long-context processing, reinforcement learning, and asynchronous training, with more environments, task diversity, and compute. Its practical appeal is a combination of broad context capacity and stronger performance on sustained software-engineering and agent-style tasks.

According to the reported evaluations, GLM-5.3 improves on its predecessor by 50% on Z.ai’s internal Code Bench and reaches open-source state-of-the-art results on Terminal Bench 3.0 and Agents’ Last Exam. The model also shows unusually strong cyber performance, including state-of-the-art CyberGym results for vulnerability discovery and more than double the predecessor’s scores on higher-level exploitation benchmarks. These results make it particularly relevant for teams working on advanced coding agents, extended investigations, and security research.

AIHubMixglm-5.3glm

Open provider page Provider docs

Quick Info

Provider: AIHubMix
Model key: glm-5.3
Release date: Aug 14, 2026
Last updated: Aug 14, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $1.1268
Output token cost: $3.9438

Limits

Output tokens: 128,000 tokens
Context window: 1,000,000 tokens

Transparent token rates

Compare GLM-5.3 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about GLM-5.3

AIHubMix

CoveragePreview

Interconnects (Sep 8, 2026) covers GLM-5.3 as part of its "Latest open artifacts (#24)" roundup, placing Zhipu's release in the context of a 2026 trend in which Chinese frontier model makers are becoming more restrictive with licenses even as Western model makers (Google, Meta) have moved to Apache 2.0. The piece highl The post directly quotes the new GLM-5.3 license clause: if the Licensee or any of its affiliates operates a Model-as-a-Service business and aggregate revenue with affiliates exceeds $10 billion USD (or equivalent) over any consecutive 12 months, the Licensee must pass Z.ai's security review before using the software o

AIHubMix

CoverageBenchmark

GLM-5.3, announced on August 14, 2026, is described by Z.ai (Zhipu AI) as a post-training-only release over the same base model as GLM-5.2, meaning "every gain comes from post-training." Z.ai frames it as "Frontier Coding with Emergent Cyber Capabilities," claiming a 50% coding improvement over GLM-5.2 on the in-house The model is text-only, always-reasoning, with a 1M-token context, 128K max output, and three thinking-effort levels (low, high, max, default max); thinking can no longer be disabled. Notably, GLM-5.3 ships under a new custom license (not MIT like GLM-5.2) that requires any Licensee operating a Model-as-a-Service busin