GLM Coding Plan

Short answer. GLM Coding Plan is Zhipu AI’s coding-agent subscription: one plan that powers ZCode (Zhipu’s own client), AutoClaw (office agent) and 20+ third-party coding tools including Claude Code, Codex, Cursor and OpenClaw, running on GLM-5.3 / GLM-5.3-Flash with 1M-token context.

Key facts.

• The same subscription unlocks GLM models across Zhipu’s own tools (ZCode, AutoClaw) and third-party coding agents — the plan’s defining characteristic.
• China tiers: Lite ¥118 / Pro ¥538 / Max ¥1,078 per month, with per-5-hour credits (2,000/12,000/28,000) and weekly credits (10,000/60,000/140,000); team seats at ¥598/¥1,198 per month. International Z.AI from $18/month.
• Time-based pricing: off-peak is 50% credit rate; peak (Mon–Fri 14:00–18:00 UTC+8) costs 2x credits.
• Runs on GLM-5.3 and GLM-5.3-Flash, both 1M-token context, with an Agentic Engineering plan-implement-iterate workflow.
• Valid only inside officially supported coding tools; self-built apps and SaaS must use the standard API.

What this means. GLM Coding Plan inverts the usual agent-vs-model relationship: instead of shipping its own coding agent as the primary surface, Zhipu makes its models available inside the tools developers already use, turning Claude Code, Codex, Cursor and OpenClaw into distribution channels for GLM-5.3 spend.

What is uncertain. The plan’s launch date is not publicly disclosed (the plan revision is dated 2026-07-30), and there is no public GitHub repository (verified against THUDM on 2026-09-27). Whether the 20+ third-party integrations stay current as those tools evolve is an ongoing maintenance question, not a static guarantee.

Sources.

evidence_id	source_name	source_url	source_type	published	verified	confidence	conflict
src-agents-glm-coding-plan-1	GLM Coding Plan landing page	https://bigmodel.cn/glm-coding	Official	—	2026-09-20	high	—
src-agents-glm-coding-plan-2	BigModel Coding Plan docs - overview	https://docs.bigmodel.cn/cn/coding-plan/overview.md	Official documentation	—	2026-09-20	high	—
src-agents-glm-coding-plan-3	Z.AI Coding Plan international docs	https://docs.z.ai/devpack/overview.md	Official documentation	—	2026-09-20	high	—

Why it matters

GLM Coding Plan is the clearest example of a Chinese model vendor monetizing its models through third-party coding tools rather than competing with them. The plan’s value is inseparable from GLM-5.3’s capability — the 1M-token context and reasoning are what those external tools actually consume — while the pricing structure (5-hour and weekly credit windows, peak 2x, off-peak half) is a consumption-control layer on top of the model.

China AI Hub analysis indicates the plan is Zhipu’s distribution strategy: it monetizes GLM models by embedding them in the coding-agent ecosystem rather than by winning the client war, and the pricing’s peak/off-peak split is an explicit capacity-management lever — the 2x peak multiplier steers usage off business hours, mirroring the off-peak economics DeepSeek introduced on its API. This is a demand-shaping mechanism, not a neutral price list.

How it differs from Kimi Code and Qwen Code

The three coding agents occupy distinct positions on the open-versus-closed and own-tool-versus-ecosystem axes.

• Kimi Code is Moonshot’s open-source (MIT) terminal agent — a first-party client bound to Kimi K3/K2.7-Code, monetized through Kimi membership and a pay-as-you-go API. The tool is free; the models are Moonshot’s own.
• Qwen Code is Alibaba’s open-source (Apache-2.0) multi-protocol agent — the CLI is free, and it routes to any provider (OpenAI, Anthropic, Gemini, Qwen, DeepSeek, Kimi, MiniMax, local). Alibaba monetizes through its Coding Plan and Token Plan, but the agent itself stays model-agnostic.
• GLM Coding Plan is not a client at all — it is a subscription that injects GLM models into other clients. Zhipu ships ZCode as a first-party option but makes the plan’s core value proposition “your existing tool, GLM’s model.”

China AI Hub analysis: the three represent three different answers to “how does a Chinese model vendor capture coding-agent demand.” Moonshot binds a free open client to its own models; Alibaba gives away an open multi-protocol client and charges for access; Zhipu skips the client war and charges for model access inside everyone else’s client. A developer already on Claude Code or Cursor faces the lowest switching cost with GLM Coding Plan — that is precisely its bet.

Practical implications

For developers. The plan’s headline value is running GLM-5.3’s 1M context and reasoning inside a tool you already know. The cost is a quota system with two refresh cadences (5-hour and weekly) and no spillover billing — exhaust a window and you wait, rather than automatically paying overage.

For cost-sensitive teams. Off-peak usage halves credit cost, and the ZCode client’s night campaign (23:00–09:00) makes GLM-5.3-Flash effectively unlimited — a concrete incentive to schedule batch or background work off-hours. Peak hours (14:00–18:00) doubling is the counterweight.

For self-builders. The plan is deliberately walled to officially supported tools; anyone building a custom app or SaaS must fall back to the standard per-token API, so the subscription does not substitute for API access.

What the evidence shows

The evidence is strong on the pricing and routing mechanics but thin on the model’s independent quality. The landing page and docs document the tier structure (¥118/¥538/¥1,078), the credit windows, the peak/off-peak multipliers, the ZCode/AutoClaw/third-party coverage, and the “valid only in supported tools” constraint; the Z.AI docs mirror the international plan from $18/month.

The gaps are launch date (not disclosed; revision dated 2026-07-30) and independent coding benchmarks for GLM-5.3 (the site records vendor-reported scores but no independent evaluation for this plan specifically). China AI Hub analysis indicates the plan’s substance is its business model — a model-vendor-as-backend to third-party agents — more than any single capability claim, and its real risk is integration maintenance: a subscription whose value rides on staying compatible with two dozen external tools inherits those tools’ release churn as an operational cost.

Where this fits

Workload	Relevance
Using GLM models inside existing tools (Claude Code, Codex, Cursor, OpenClaw)	High
Agentic Engineering (plan-implement-iterate, 1M context)	High
ZCode first-party client / AutoClaw office agent	High
Off-peak / scheduled batch work (night campaign)	High
Self-built apps / SaaS on the standard API	Low (plan not valid there)
Independent benchmark evaluation of GLM-5.3 coding	No evidence recorded

Relevance judgments are China AI Hub analysis based on documented capabilities, not vendor claims.

Field reference

Field	Value	Evidence type
Underlying model	GLM-5.3, GLM-5.3-Flash	Official
Target users	Developers using coding tools	Official
Platform	Coding-agent subscription (ZCode, AutoClaw, 20+ third-party tools)	Official
OS	Not publicly documented (via supported coding tools)	Not publicly documented
Browser / computer use	Not publicly documented	Not publicly documented
Coding	Yes (coding-agent subscription)	Official
Autonomous task execution	Yes (Agentic Engineering plan-implement-iterate)	Vendor-reported
MCP	Yes	Official
Tool calling	Yes	Official
Memory	Yes	Official
Workflow	Plan-implement-iterate; credits refresh per 5-hour window and weekly	Official
API	Yes	Official
Pricing	China ¥118/¥538/¥1,078 per month; international from $18/month	Vendor-reported
Region	China (bigmodel.cn) and international (Z.AI)	Official
Open-source	No — proprietary	Official
Deployment	Cloud	Official
Limitations	Quota caps; valid only in supported tools; peak 2x credits	Official
Source	BigModel Coding Plan	Official
Last verified	2026-09-20	Official

See the Zhipu AI company profile, the GLM-5.3 and GLM-5.3-Flash models, the choosing-a-coding-model guide, and the site’s AI agents and tool calling technology pages.

Labels used above: Official fact (from the BigModel Coding Plan and Z.AI docs), Vendor-reported claim (pricing and capability statements by Zhipu AI), and China AI Hub analysis (our synthesis, always introduced as such).