GLM-5.3

Z.ai's official blog confirms GLM-5.3 launched on August 14, 2026, as a post-training-only release built on the same base model as GLM-5.2, with the IndexShare, SAO, and slime stack handling long-horizon training. The post states GLM-5.3 achieves a 50% improvement over GLM-5.2 on Z.ai's in-house Code Bench and claims o The blog details that GLM-5.3's cyber capabilities emerged faster than expected during post-training scaling, with gains more than doubling GLM-5.2 on exploitation benchmarks like ExploitGym and ExploitBench. A comparison table shows specific scores (Terminal Bench 3.0: 28.3 vs 4.6; CyberGym: 84.5 vs 77.2; ExploitBench

BetaNews reports that Z.ai announced GLM-5.3 on August 14, 2026, alongside OpenVuln, a scanning service internally branded as "VulnHunter" that uses GLM-5.3 to check public code repositories for security flaws, with Z.ai publishing aggregate security scores while keeping detailed findings private until fixes are ready. According to the article, Z.ai's public disclosure ledger credits GLM-5.3 with 2,436 vulnerability findings across 269 open-source projects, including 1,097 rated critical or high severity, with several flaws found in the Linux kernel and widely used VMware and Apache projects. Z.ai stated it would hold back GLM-5.3's

Morph's August 21, 2026 third-party technical summary corroborates Z.ai's GLM-5.3 launch narrative and adds developer-relevant details: API pricing is unchanged from GLM-5.2 at $1.40 per million input tokens and $4.40 per million output tokens, with a 1M-token context window, 128K max output, and always-on reasoning ac Independent Artificial Analysis evaluation cited on the page gives GLM-5.3 an Intelligence Index score of 60, ranking 8th out of 182 models and tying Kimi K3 at roughly one-fifth the price. The post frames the 4.6 to 28.3 Terminal-Bench 3.0 jump and 46.2 to 66.9 DeepSWE v1.1 gain as the headline agentic-coding improvem

According to InferenceX's technical analysis, GLM-5.3 was announced on August 14, 2026, and is a post-training-only release that uses the same base model as GLM-5.2, meaning all capability gains come from post-training work on the IndexShare, SAO, and slime stack. The page explicitly references Z.ai's GLM-5.3 blog and The same analysis positions GLM-5.3 as the highest-ranked open-weights model on long-horizon coding benchmarks, with Z.ai claiming a 50% coding improvement on its in-house Code Bench over GLM-5.2 and open-source SOTA on Terminal-Bench 3.0 and Agents' Last Exam (CLI). Z.ai also reports emergent cyber capability, includi