GLM (Z.ai / Qingyan) Review: The Coding Model That Costs 1/40th of Its Rival

Tools Updated 2026-11-04 · 3 min read · #chatbot #coding #agents #api

Status board (checked 2026-11-04)


Who made it	Zhipu / Z. AI (Beijing)
Entry points	chat.z.ai (global) · chatglm.cn 清言 (domestic) · bigmodel.cn / docs.z.ai (API) — accounts and billing don’t mix across the three
Export-control status	✅/⚠️ GLM-5.3 (custom license) and GLM-5.3-Flash (MIT) weights are already public, which the draft excludes; no public export-registration announcement from Zhipu found (MiniMax and DeepSeek have one); draft comments close Nov 15

What it is

The assistant side of the lab behind GLM-5.3-Flash. The picker offers two gears: GLM-5.3, the flagship (launched Aug 14; Z.ai claims a 50% coding gain over GLM-5.2 on its own Code Bench, 1M-token context, text-only — and its open weights ship under a custom license, not MIT), and GLM-5.3-Flash, natively multimodal — its pitch is “capabilities aligned with Claude Opus 4.8 at ~1/40th the price.” The base chat is free; domestic Qingyan meters heavy features (PPT agent, video, HD drawing) with points, and its VIP/SVIP (¥79/¥229 monthly, domestic only) buy quota multipliers, not feature unlocks.

Free tier

Web chat basics — dialogue, web search, document analysis — cost nothing; heavy agent/video tasks eat points on the domestic front.
The most-searched free line: GLM-4.7-Flash API is permanently free (200K context), and new API accounts get tens of millions of trial tokens (the current promo: 8M tokens/day for the first 5 days).

Getting started (3 steps)

Open chat.z.ai (email or phone) for the global version, or chatglm.cn for the domestic one; pick GLM-5.3 for hard reasoning, Flash for everyday speed.
Coding agents: subscribe to the GLM Coding Plan (domestic Lite ¥118/mo) and point Claude Code / Codex / OpenCode at api.z.ai/api/anthropic — and know the trick: off-peak calls (outside 14:00–18:00 Beijing) cost half the plan credits.
Spot-check like always: years, numbers, quotes — and remember the flagship is text-only; for pictures hand it to Flash.

Best use

Scenario	Why GLM wins here
Claude Code on a budget	Anthropic-compatible endpoint + Coding Plan = flat-rate agentic coding; API equivalent: $1.4/$4.4 per M (flagship) vs $0.15/$0.50 (Flash)
Long Chinese docs	Home-town Chinese: public documents, weekly reports, CN research papers read better than English-first models
Try-before-you-buy	The exact models are open weights — graduate to running them at home when the bill grows

Honest weaknesses

Knowledge cutoff around end of 2025 — fresh facts lean on the web-search switch.
Generated UIs are “usable, not beautiful”: heavy visual design isn’t its turf.
Three products, three billing worlds (global chat / domestic points / API prepaid) — mixing them up is the #1 beginner bill surprise.

FAQ

Chat.z.ai or Qingyan — which do I use?

Global users: chat.z.ai, email signup, English fine. Domestic: chatglm.cn (phone/WeChat login, points system). Different accounts, no shared history.

Is the Coding Plan worth it vs plain API?

For daily agent coding, usually yes — the Lite plan’s credits run far, especially with the 50% off-peak discount (peak window: 14:00–18:00 Beijing, everything outside it is half-credits). For occasional calls, pay-as-you-go Flash is cheaper than any subscription.

The open model and this chat — same brain?

Yes: the hosted flagship is the same GLM line whose weights you can run on your own rig; this page’s status box tracks the export-catalogue draft that could change overseas commercial paperwork.

📌 Summarised from official docs and hands-on samples. Product features and pricing change often — official site wins.

← Newer: Doubao Review: The Lowest-Pressure AI to Just *Start* Using Older: Kimi Review: Throw a Whole Textbook at It →