GLM-5.2

GLM-5.2 is an open-weights large language model from z.AI designed for long-running coding sessions and agentic workflows. Coverage from mid-2026 describes it as an open-source Chinese release that quickly drew notice from Silicon Valley practitioners, with quoted reactions highlighting surprisingly strong coding behavior for a model of this lineage. The Business Insider write-up frames GLM-5.2 as a serious new entry in the open-weight space, comparable in ambition to other recent flagship releases.

The GLM-5.2 base model also serves as the foundation for its successor, with later post-training producing the GLM-5.3 family and measurable gains over GLM-5.2 on internal code benchmarks plus Terminal Bench 3.0, Agents' Last Exam, and CyberGym vulnerability-discovery tasks. That lineage positions GLM-5.2 as a capable general base for developers who want an open-weight backbone suitable for code generation, agent pipelines, and security research prototypes, while teams needing the very strongest long-horizon coding or cyber performance can step up to the post-trained variants built on the same base.

Pioneerzai-org/GLM-5.2-Fastglm

Open provider page Provider docs

Quick Info

Provider: Pioneer
Model key: zai-org/GLM-5.2-Fast
Release date: Jun 13, 2026
Last updated: Jun 13, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $2.10
Output token cost: $6.60

Limits

Output tokens: 128,000 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

Pioneer

CoverageBenchmark

The Baidu Cloud blog (published September 7, 2026) explicitly names GLM-5.2 and reports an official June 15, 2026 launch with three evolutionary milestones: continuation of the Mixture-of-Experts sparse architecture at 744B total parameters (40B activated), a context window expansion from 200K to 1M tokens, and an open The same article describes GLM-5.2's MoE implementation as comprising 64 specialized subnetworks with dynamic gating that activates 2–4 experts per token, claims a 60% reduction in theoretical FLOPs and inference latency under 300ms on domestic GPU clusters, and frames the 1M-token context expansion as enabling long-do

Pioneer

CoverageBenchmark

MarkTechPost's July 2026 comparison directly names Z.ai's GLM-5.2 as a sparse Mixture-of-Experts model with 744B total parameters and 40B active parameters, a 1M-token context window (with up to 131K output), and a text modality. The article confirms a mid-June 2026 release by Zhipu AI (Z.ai) and positions GLM-5.2 alon The MarkTechPost piece also notes that GLM-5.2 is the smallest of the three trillion-scale MoE contenders by total parameters, and that it earned its place in the comparison because it led the open-weight field before Kimi K3 shipped on July 16, 2026. License, context, and modality attributes are tabulated explicitly f