GLM-5.2 — Model availability¶

Development line: project:glm-5-2 · thread model-availability
Last event: 2026-06-14 · 1 dated since 2026-06-14 · Researched: 2026-09-05 · confidence: medium

What it is¶

GLM-5.2 is a still-listed Z.ai model for software-engineering agents.

• Reasoning
• Streaming
• Function calls
• Context caching
• Structured JSON
• MCP

1M context and 128K output.

Retain GLM-5.2 for compatibility or a controlled comparison, but use GLM-5.3 for new coding work.

Development line¶

• 2026-06-14 — Z.ai documentation identified GLM-5.2 as its latest model. On 2026-06-14, Z.ai linked DevPack documentation for the latest model under the GLM-5.2 project line. This marks an availability milestone, but we did not research the post or documentation. We assert no release details, capabilities, or access terms.

What changed¶

• 2026-06-14 — The linked “latest model” URL points to a GLM-5.3 switching guide, so it does not prove a GLM-5.2 launch.
• 2026-06-16 — Z.ai released GLM-5.2 and open-sourced it under MIT, adding a 1M lossless context model for long-horizon coding tasks.
• 2026-07-08 — An independent CAISI assessment added a safety caveat: deployment safeguards did not prevent agentic exploit-development assistance, and operators can remove safeguards from self-hosted open weights.
• 2026-08-18 — GLM-5.3 superseded GLM-5.2 as Z.ai’s latest flagship, with a vendor-reported 50% coding gain on Z.ai Code Bench.

How to use this¶

As of 2026-06-14, check Z.ai’s DevPack latest-model documentation when selecting GLM-5.2. Do not assume an earlier GLM model remains the current documented option.

1. Create a Z.ai API client and send a Bearer-authenticated request to https://api.z.ai/api/paas/v4/chat/completions with model: "glm-5.2" and at least one user message. — https://docs.z.ai/guides/llm/glm-5.2
2. For a difficult coding task, enable thinking and set reasoning_effort to max. Cap max_tokens to the output actually required. — https://docs.z.ai/guides/llm/glm-5.2
3. Use stream: true when the caller can consume Server-Sent Event chunks. Omit it or set it to false for a synchronous response. — https://docs.z.ai/api-reference/llm/chat-completion
4. Treat returned function-call arguments as untrusted input. Validate them against the declared schema before execution. — https://docs.z.ai/api-reference/llm/chat-completion

Best practices¶

• Give a refactoring task explicit business-logic, API, runtime, dependency, and verification boundaries. Require a plan and relevant test results before accepting the change. — https://docs.z.ai/guides/llm/glm-5.2
• Keep thinking.clear_thinking enabled for ordinary chat. Preserve prior reasoning only when the complete, unmodified, correctly ordered history is available. — https://docs.z.ai/api-reference/llm/chat-completion
• Evaluate GLM-5.3 first for new coding deployments. Keep GLM-5.2 only where its model ID is an intentional compatibility constraint. — https://docs.z.ai/guides/llm/glm-5.3

Superseded by this¶

• :

Still unknown¶

• The 2026-06-14 item predates Z.ai’s confirmed 2026-06-16 release. Available first-party material confirms advance Coding Plan access before release but does not date that preview to June 14.
• Z.ai’s official pages disagree on GLM-5.3 timing. The research index shows 2026-08-14, while release notes show 2026-08-18; we use the latter for the supersession date.
• We made no authenticated API request. Account-specific availability, quota, and regional access remain unverified.

Sources¶

Agent brief¶

• Subject: project:glm-5-2, thread model-availability, 1 dated events 2026-06-14 → 2026-06-14.
• Practical note: As of 2026-06-14, check Z.ai’s DevPack latest-model documentation when selecting GLM-5.2. Do not assume an earlier GLM model remains the current documented option.
• Confidence: medium. Dated supersedes above are the authority for what is obsolete.