GLM-5.2

A Baidu Cloud blog post dated September 7, 2026 describes GLM-5.2 as an open-source coding model released to all users on June 13, 2026, presenting technical architecture, performance optimization, and engineering details. The post claims GLM-5.2 maintains the GLM-5 series' super-large parameter design with 744 billion The reported optimizations include a 67% reduction in attention computation while maintaining 98% inference accuracy on the described tests, and only an 18% increase in memory usage when handling long contexts. The article frames GLM-5.2's positioning around long-context reasoning and multi-language modeling for comple

The NIST Center for AI Standards and Innovation (CAISI) published an official assessment of Z.ai's GLM-5.2 on July 8, 2026, with the PDF dated July 17, 2026. The report documents that GLM-5.2 was released as an open-weight model by the PRC-based company Z.ai (formerly Zhipu AI) on June 16, 2026. CAISI's independent eva On safeguards and security, CAISI's evaluation reports mixed results: GLM-5.2's safeguards allow assistance with agentic cyber exploit development, block fewer sensitive biological questions than reference U.S. models, yet appear potentially more robust against agent hijacking and jailbreaking attacks than other evalua

A Business Insider Facebook post from June 22, 2026 reports that GLM-5.2, a new open-source Chinese AI model focused on long-form coding tasks, is drawing attention from Silicon Valley's AI elite. The post frames GLM-5.2 as an open-source release positioned for extended coding workflows, crediting Z.ai (formerly Zhipu The candidate does confirm that GLM-5.2 exists as a named, distinct version within the GLM family and that it is positioned for long-form coding — facts corroborated by the NIST/CAISI assessment and the Semgrep cyber benchmark candidates. However, the supplied page excerpt provides no benchmarks, no parameter counts, n

Semgrep published a technical blog on June 22, 2026 titled "We have Mythos at Home: GLM 5.2 beats Claude in our Cyber Benchmarks," reporting results from running GLM 5.2 against their IDOR benchmark — the same dataset and prompt they use to evaluate frontier coding agents. The post is authored by Semgrep security resea The candidate explicitly names GLM 5.2 (matching the exact model subject) and attributes evaluation to an independent third-party security vendor, not to Requesty as a serving provider, keeping the focus on the model itself rather than gateway concerns. The blog post is scoped to GLM 5.2 specifically, not to a sibling