GLM-5.2

GLM-5.2 is Z.ai's flagship foundation model, purpose-built for the era of long-horizon tasks - tasks stretched over many hours of work, covering full engineering cycles from requirements to deployment. It is the direct successor to GLM-5.1 and the second model in the GLM-5 line, continuing Z.ai's positioning as the open-source leader in agentic software engineering. The model powers the z.ai homepage (since launch it carries the tagline 'Advanced AI Chatbot & Agent powered by GLM-5.2').

Key quality leap: usable 1M context (5x larger than GLM-5's 200K). Z.ai emphasises that the foundation of long-horizon tasks is not simply having a 1M window, but sustaining stable model performance across ultra-long context. GLM-5.2 underwent months of specialised training for large-scale Coding Agent scenarios - and in selected real-world benchmarks it even surpasses Claude Opus. Maximum output remains at 128K tokens, modalities: text-in / text-out.

New capabilities over GLM-5: native MCP (Model Context Protocol) support - flexible integration with external tool servers and data sources. The reasoning_effort parameter lets you tune reasoning depth (default 'max'). The model retains module boundaries, architectural constraints, API contracts, directory structures, and historical project decisions within a single reasoning workflow - critical for project-level codebase takeover.

Benchmarks (open-source SOTA): Terminal-Bench 2.1: 81.0 (leap from GLM-5.1's 62.0, only 4 points below Claude Opus 4.8's 85.0), SWE-bench Pro: 62.1 (up from GLM-5.1's 58.4), FrontierSWE: within 1% of Opus 4.8, outperforms GPT-5.5 and Opus 4.7 on multiple benchmarks. The model is the top-ranked open-source result across all three: FrontierSWE, PostTrainBench, SWE-Marathon. GLM Coding Plan developers (early access before release) reported especially: stronger project-level context capacity, more stable long-horizon task execution, better adherence to production-grade engineering standards (lint, tests, commit conventions), and stronger client-side and mobile engineering (Android Kotlin + ADB/logcat, WeChat Mini Programs, Remotion Code-to-Video).

Availability: API at api.z.ai (model name: glm-5.2, OpenAI-compatible format at api.z.ai/api/paas/v4/), chat at z.ai (main production model), open-weights for self-hosting. GLM Coding Plan Pro/Max subscription grants model access in Claude Code, Open Code, and other tools. SDKs: zai-sdk (Python), ai.z.openapi:zai-sdk (Java), plus OpenAI Python SDK support. Z.ai-documented use cases: project-level codebase takeover, long-horizon refactoring, production-grade standards stress test, mobile on-device debugging loop, WeChat Mini Program development, mini game development, research reproduction from paper to runnable code, code-to-video via Remotion.