GLM-5.2

GLM-5.2 is positioned by Z.ai as their latest flagship model built specifically for long-horizon tasks, representing a substantial capability leap over its predecessor GLM-5.1. The release introduces what the official blog describes as a solid the cataloged API limit that can stably sustain extended agentic work, marking the first time this family has combined long-horizon competence with such an extended context. For developers, this means the model is aimed at workflows where an agent must remain coherent across very long, messy coding trajectories rather than just processing more tokens in isolation.

The model is distributed openly, with weights published on Hugging Face under the zai-org organization and accompanying implementation code in the zai-org/GLM-5 GitHub repository under an MIT license, and it is accessible through the Z.ai API as well as on the Z.ai Coding Plan. Z.ai highlights stronger coding capabilities supported by multiple thinking-effort levels, letting users tune the trade-off between response latency and task performance. An architectural refinement called IndexShare is introduced to reuse the same indexer across groups of sparse attention layers, and the multi-token prediction layer has been tuned to extend speculative-decoding acceptance length, both targeting efficiency at very long context lengths.

SCNet Token PlanGLM-5.2glm

Open provider page Provider docs

Quick Info

Provider: SCNet Token Plan
Model key: GLM-5.2
Release date: Jun 13, 2026
Last updated: Jun 13, 2026
Input modalities
Output modalities
Capabilities

Cost

A provider subscription or plan supersedes token-based pricing for this model.

Limits

Output tokens: 131,072 tokens
Context window: 1,000,000 tokens

Latest news about GLM-5.2

SCNet Token Plan

Coverage

SaferAI published the first European independent evaluation of GLM-5.2 on August 2, 2026, testing Z.ai's open-weight flagship (released June 16, 2026) across the four systemic risk areas defined in the EU General-Purpose AI Code of Practice: Loss of Control, Cyber Offense, CBRN, and Harmful Manipulation. The evaluation On specific risk dimensions, GLM-5.2 was approximately level with Opus 4.7 and slightly below GPT-5.5 on biological knowledge (about two months behind the frontier), around two to four months behind on cyber (at the level of Opus 4.6 and near GPT-5.5), and furthest behind on software engineering (below Opus 4.6 and GPT

SCNet Token Plan

Coverage

The U.S. National Institute of Standards and Technology's Center for AI Standards and Innovation (CAISI) published an independent assessment of Z.ai's open-weight GLM-5.2 model on July 17, 2026, based on evaluations conducted around the model's June 16, 2026 release. According to CAISI's results, GLM-5.2 was likely the On safeguards and agent security, CAISI found mixed results: GLM-5.2's safeguards permit assistance with agentic cyber exploit development and block fewer sensitive biological questions than reference U.S. models, but the model appears potentially more robust against agent hijacking and prompt-based jailbreaking attack

SCNet Token Plan

CoverageAnalysis

A Hacker News discussion thread (filed around June 17, 2026) aggregates community reaction to Artificial Analysis identifying GLM-5.2 as the new leading open-weights model. Practitioners compare GLM-5.2's reasoning and output behavior to frontier systems: commenters report that GLM-5.2 "Max" effort mirrors Opus 4.8 "Ma The evidence is community-reported and anecdotal (token counts and behavior impressions from individual users), not a primary benchmark, and the thread will age quickly — observations are anchored to mid-June 2026, months before the current date. Attribution is to Z.ai's GLM-5.2 as the model subject; the SCNet Token Pl

SCNet Token Plan

CoverageBenchmark

Semgrep's security research team published benchmark results on June 22, 2026 showing that GLM 5.2 outperformed Claude Opus 4.8 on their IDOR (Insecure Direct Object Reference) benchmark when given nothing but a single prompt. The test used the same dataset and prompt Semgrep employs to evaluate frontier coding agents, The finding positions GLM 5.2 as a competitive open-weight option for prompt-only cyber/code-security workflows, beating the named Claude frontier comparator on this specific benchmark. Semgrep notes it ran the open-source models against the same IDOR benchmark used for its frontier coding-agent evaluations, suggesting