GLM 5.3

Z.ai released GLM-5.3 on August 14, 2026 as a post-training-only upgrade of the same 753B mixture-of-experts base that powers GLM-5.2—no new pretraining was performed. All reported gains come from scaled post-training targeting long-horizon agentic coding and vulnerability research: Terminal-Bench 3.0 jumps from 4.6 to API pricing for GLM-5.3 is unchanged from GLM-5.2 at $1.40 per million input tokens and $4.40 per million output tokens. On the Artificial Analysis Intelligence Index, the model scores 60 and ranks 8 of 182, reportedly tying Moonshot AI's Kimi K3 at roughly one-fifth the price. A public ModelOpt NVFP4 expert quantizati

Z.ai has released GLM-5.3, an update to its large language model that targets complex coding and long-horizon agentic tasks. According to MarkTechPost's coverage of the release, GLM-5.3 runs on the same 743B base model as the previous GLM-5.2, with every reported improvement coming from scaled post-training — more task The reported gains land most heavily on the longest-horizon benchmarks. Terminal-Bench 3.0, which measures complex coding tasks, jumped from 4.6 to 28.3 in the new version. Cybersecurity performance also moved further than Z.ai expected, with CyberGym reaching a score of approximately 84. For developers evaluating GLM

Together AI's documentation for GLM-5.3 confirms the hosted endpoint identifier `zai-org/GLM-5.3`—matching the subject's model key—and lists a 1M-token context window carried over from the GLM-5.2 base. The architecture employs Z.ai's IndexShare design for efficient long-context processing alongside SAO with compaction Capability highlights from Together AI's listing describe frontier agentic coding trained on environments modeled after real production workflows—including multi-step dependencies and hidden state—enabling the model to own multi-day engineering tasks end-to-end. Z.ai also directed post-training toward vulnerability dis