GLM-5.2: Z.ai's flagship for coding and stable agents

GLM-5.2 is Z.ai's latest flagship (2026) and the top of the GLM line in Council's lineup. Its profile is narrow and honest: strong coding and — the trait Z.ai leads with — stable multi-step agent execution, meaning agent loops that complete without the mid-run derailments cheaper models are prone to. It ships a 203K-token context window and is text-only: no vision input, no native web search. Pricing is $1.40 per million input tokens and $4.40 per million output, putting it in the value-flagship bracket — under a third of Claude Sonnet 5's standard output price for a model that competes credibly on contained coding tasks. Versus its open-weight predecessor GLM-5 (released February 11, 2026 at $0.60/$1.92 via OpenRouter), GLM-5.2 is the more capable coder and the steadier agent for a bit over double the price; GLM-5 remains the budget seat. The case for GLM-5.2: agent pipelines where reliability-per-dollar matters, and bulk coding work a tier below the frontier flagships. The case against: multimodal tasks, time-sensitive questions, and whole-repo jobs beyond 203K tokens.

Specs

Best at

• Stable agent execution — the differentiator: multi-step tool runs finish without drifting, at a price where you can run many of them.
• Contained coding tasks — solid codegen and edits when the spec is clear and the scope fits 203K tokens.
• Reliability-per-dollar — $4.40/1M output for agent work that would cost 3–5× on a Western flagship.
• Council diversity — a strong Chinese-lab voice with failure modes uncorrelated with OpenAI/Anthropic/Google.

Where it loses

• No vision, no search — text-only and offline; screenshots and current events need other seats.
• Context ceiling — 203K is a fifth of the 1M-class flagships; large monorepos won't fit.
• Frontier reasoning — it competes a tier below GPT-5.6 Sol, Claude Fable 5, and Grok 4.5 on hard problems.

GLM-5.2 vs GLM-5

GLM-5 (released February 11, 2026) is Z.ai's open-weight foundation model — production-grade large-scale programming and long-horizon agent workflows at $0.60/$1.92 via OpenRouter, with the same 203K context window. GLM-5.2 keeps the window and the coding-plus-agents identity but raises capability and execution stability for roughly 2.3× the price. The split is clean: GLM-5 for high-volume pipelines where cost dominates, GLM-5.2 when a failed agent run costs more than the tokens saved. Both are in Council's active lineup.

How Council AI runs it

GLM-5.2 is available on every paid Council AI plan (Plus, Pro, Ultra), routed through OpenRouter — Z.ai has no dedicated provider integration, and doesn't need one. In a council it answers in parallel with models from up to 8 other labs while the moderator synthesizes a consensus response with an agreement score. Its price makes it an easy extra seat: adding GLM-5.2 to a council barely moves the cost while adding an independently-trained perspective, which is exactly what makes agreement scores meaningful.

Frequently asked questions