{"id":17430,"date":"2026-08-17T22:28:18","date_gmt":"2026-08-17T22:28:18","guid":{"rendered":"https:\/\/nile1.com\/en\/?p=17430"},"modified":"2026-08-17T22:28:26","modified_gmt":"2026-08-17T22:28:26","slug":"z-ai-debuts-743-billion-parameter-coding-model-glm-5-3-ahead-of-open-weight-release","status":"publish","type":"post","link":"https:\/\/nile1.com\/en\/2026\/08\/17\/z-ai-debuts-743-billion-parameter-coding-model-glm-5-3-ahead-of-open-weight-release\/","title":{"rendered":"Z.ai Debuts 743-Billion Parameter Coding Model GLM-5.3 Ahead of Open-Weight Release"},"content":{"rendered":"

Safety evaluations will precede API access and open-weight releases for GLM-5.3<\/a>, a 743-billion-parameter programming model launched Thursday by Chinese artificial intelligence lab Z.ai<\/a>. Positioned by the developer as the top-performing open-weights coding model available, the system is currently accessible through the ZCode platform and GLM Coding Plan subscriptions.<\/p>\n

“Scaling post-training is all we did for GLM-5.3,” the company wrote in its launch post. “With GLM-5.2 we built the stack… Over the past month we kept scaling on this stack: more environments, more diverse tasks, and more compute spent training on them.”<\/p>\n

Development prioritized token efficiency rather than sheer size dominance. Built on a 743-billion-parameter architecture, GLM-5.3 consumes significantly fewer tokens per task compared to its predecessor. In machine learning systems, parameters act as the adjustable controls handling information processing, while tokens represent the basic units of data consumed or generated by a model.<\/p>\n

Efficient token consumption directly translates to lower operational overhead during extended programming sessions, where models must repeatedly digest codebases and generate iterative solutions without exceeding context memory or escalating computing costs.<\/p>\n

Internal evaluations published by Z.ai show GLM-5.3 achieving a 34.5% success rate on its in-house Z.ai Code Bench at Max effort while burning roughly 75,000 output tokens per task, outperforming GLM-5.2’s 23.4% mark at 96,000. When measured against proprietary rivals, the company noted that while the model surpasses Claude Opus 4.8<\/a> on token economy, it “remains behind Claude Fable 5, which reaches 39.5% at Max effort.”<\/p>\n

On primary programming benchmarks, GLM-5.3 delivers strong results, outperforming domestic competitor Kimi K3 across major evaluation metrics.<\/p>\n

In Terminal Bench 3.0 evaluations\u2014which measure automated shell and tool interactions inside real Linux environments\u2014GLM-5.3 registered 28.3, trailing closed models Fable 5 at 33.7 and GPT-5.6 Sol at 34.6. On DeepSWE v1.1, a benchmark for fixing real GitHub issues end-to-end, open rival Kimi K3 at 67.5 and Fable 5 at 69.7 both beat GLM-5.3’s 66.9 score.<\/p>\n

The performance pattern indicates that while GLM-5.3 clears its own predecessor and some open peers, closed U.S. models still lead the headline coding boards.<\/p>\n

Security testing revealed another major jump, with GLM-5.3 leading CyberGym at 84.5% and more than doubling GLM-5.2 on exploitation benchmarks. According to Z.ai, the model flagged 2,436 vulnerabilities across 269 open-source projects, including 1,097 medium-to-high severity flaws.<\/p>\n

“GLM-5.3 takes agentic coding to the next level, delivering a dramatic improvement over GLM-5.2 while achieving better results with fewer output tokens,” Z.ai posted on X. “GLM-5.3 is available now through GLM Coding Plan and ZCode. API access and open weights will be released in stages following rigorous safety evaluations.”<\/p>\n

On price, the gap with U.S. frontier models remains a primary attraction for open-weights technology. Z.ai’s GLM Coding Plan operates on a points quota system with half-priced off-peak calls, while Zhipu’s API pricing sits at roughly a tenth of U.S. frontier per-token rates\u2014GLM-5.2’s official rate was $1.40 in \/ $4.40 out per million tokens, compared to $1.75 \/ $14 for GPT-5.3-Codex and high-tier costs for Claude Opus 4.8.<\/p>\n

Based in Beijing, Z.ai remains on the U.S. Entity List, prohibiting American firms from exporting controlled technology to the group. Despite those restrictions, GLM remains widely popular, with Chinese open-weight models outpacing American competitors in token usage on OpenRouter<\/a>.<\/p>\n

Full availability of downloadable software remains pending, as public release of GLM-5.3 weights is scheduled for roughly two weeks following the launch post while safety evaluations are finalized.<\/p>\n

\n

Read also:<\/h3>\n