GLM-5.2

GLM-5.2 is positioned as Zhipu AI's flagship successor to GLM-5.1, with the launch framing it explicitly as "Built for Long-Horizon Tasks." Rather than marketing raw scale alone, the model is engineered around sustaining quality across extended, multi-step coding-agent trajectories. Its headline feature is what Z.ai calls a "solid the cataloged API limit context" that "stably sustains long-horizon work," described in the developer documentation as "the cataloged API limit Lossless Context." This focus on usable long context, rather than just accepting more tokens, shapes how the model is intended to be deployed in real agent pipelines.

Architecturally, GLM-5.2 introduces IndexShare, a scheme that reuses the same indexer across every four sparse attention layers and reportedly cuts per-token FLOPs by roughly 2.9× at full context, along with an improved multi-token prediction layer that lifts speculative decoding acceptance length by up to 20%. The model also exposes multiple "thinking effort" levels so users can trade latency against reasoning depth, and it is released under a permissive MIT-style license with weights on Hugging Face and code on GitHub. Z.ai's documentation and homepage frame GLM-5.2 as achieving "Open-source SOTA Performance" in coding and agentic benchmarks, making it a natural fit for teams building autonomous coding assistants, research agents, or retrieval-heavy workflows that need sustained reasoning across very long inputs.

Zhipu AIglm-5.2glm

Open provider page Provider docs

Quick Info

Provider: Zhipu AI
Model key: glm-5.2
Release date: Jun 13, 2026
Last updated: Jun 13, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $1.40
Output token cost: $4.40

Limits

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

Transparent token rates

Compare GLM-5.2 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about GLM-5.2

Zhipu AI

Official sourceAnnouncement

Z.ai announced GLM-5.2 on June 16, 2026 as a flagship model purpose-built for long-horizon tasks, delivering a solid 1M-token context that sustains extended coding-agent workflows. The release post details a new IndexShare architecture that reuses the same indexer across every four sparse attention layers, cutting per- On long-horizon coding benchmarks reported by Z.ai, GLM-5.2 trails Opus 4.8 by only 1% on FrontierSWE while edging GPT-5.5 by 1% and Opus 4.7 by 11%, outperforms both Opus 4.7 and GPT-5.5 on PostTrainBench (ranking second to Opus 4.8), and is evaluated on the SWE-Marathon ultra-long-horizon benchmark covering compilers

Zhipu AI

Coverage

The NIST Center for AI Standards and Innovation (CAISI) published an independent assessment of Z.ai's GLM-5.2 on July 8, 2026, finding it was probably the most capable open-weight model at its June 16, 2026 release. CAISI's evaluations place GLM-5.2's overall capabilities at a level similar to GPT-5.2 (December 2025) a CAISI also reported mixed safeguards and security findings: GLM-5.2's safeguards permit assistance with agentic cyber exploit development and block fewer sensitive biological questions than reference U.S. models, but it appears potentially more robust against agent hijacking and prompt-based jailbreaking than other eva

Zhipu AI

CoverageAnalysis

A June 18, 2026 Hacker News thread (916 points, 444 comments) discussed Artificial Analysis's ranking of GLM-5.2 as the new leading open-weights model, capturing developer reaction and practical reasoning-efficiency observations. Practitioners reported GLM-5.2 Max spending roughly 45k tokens and over 15 minutes reasoni Commenters noted that running GLM-5.2 at High effort instead of Max cuts output token usage by roughly 2–2.5× with little perceived quality drop for most tasks, drawing comparisons to Opus 4.8 Max thinking behavior and suggesting GLM-5.2 can serve as a lower-cost substitute when tuned. The community signal reinforces t

Zhipu AI

CoverageBenchmark

Semgrep published a third-party cyber-benchmark evaluation on June 22, 2026 reporting that GLM-5.2, given nothing but a prompt, beat Claude Opus 4.8 on their IDOR benchmark — the same dataset and prompt the vendor uses to evaluate frontier coding agents. The result positions the open-weight GLM-5.2 as competitive with The Semgrep piece frames the finding as notable because GLM-5.2 is among the strongest open-weight options evaluated on cyber-oriented coding-agent workloads, complementing Z.ai's own long-horizon coding claims with an independent vendor benchmark. The authors — researchers from Semgrep's security team — run GLM-5.2 he