Z.ai (GLM Coding Plan)
Coding plans · reviewed by Stan K. · Source: data snapshot · data as of 2026-11-11 · scored at Pro
The only plan that publishes its credit math — 12,000 credits/5h on Pro, half-price off-peak, works in 20+ agents. Also the weakest privacy record here.
#2 of 10 in AI Coding Plans
Tiers — what each dollar buys
| Tier | $/mo | Annual | Quota & notes |
|---|---|---|---|
| Lite | $18 | — | 2,000 credits/5h · 10,000/weekly · "starting at just 18 USD per month" (official docs) |
| Pro scored | $80 | — | 12,000 credits/5h · 60,000/weekly · 3-month billing $192 (= $64/mo, −20%) |
| Max | $168 | — | 28,000 credits/5h · 140,000/weekly · prices live only on the JS-rendered subscribe page |
Agent support: 20+ agents — Widest official support in the table: Claude Code, ZCode, Codex, OpenCode, AutoClaw, OpenClaw, Cline + "20+ Agent Tools", with MCP tools included.
PlanScore breakdown — why 68.5/100
10 plan criteria × 10 points; PlanScore = mean × 10. Low pillars are visible on purpose — that's the point of a score you can audit. Methodology →
What it does well
- ▸The only provider that publishes the full credit math: multipliers (6.9/1.7/24 in/cached/out ÷10,000) — you can compute your tokens before you pay
- ▸Off-peak = 50% credit rate; US evening work lands inside the discount window by timezone luck
- ▸Works in 20+ agents officially — Claude Code, Codex, OpenCode, Cline, ZCode and more
- ▸Cheapest verified big-token volume: our own 30-day Pro anchor measured 9.63B tokens (bonus-inflated — see traps)
Where it costs you
- ▸Weakest privacy record in the table: the Sept 2026 .git silent-upload incident, training on individual plans by default, China data residency
- ▸Bonus credits inflate every hype number — the FAQ itself says reset cards expire
- ▸July 2026 credits rework burned legacy plan buyers; subscriptions non-refundable
- ▸Pro/Max prices are only on the JS-rendered subscribe page — not verifiable in static docs
Hidden traps — the stuff the pricing page doesn't say
Quota Reset Cards are official campaign bonuses with their own validity period — "they expire automatically" (devpack FAQ). Our 30-day anchor spent 473.94K credits against an official 240K/mo quota — when the promos end, so does most of that volume. Treat ~9.6B/mo as an upper bound, not the product.
Sept 2026: the ZCode client silently uploaded whole .git workspaces to Aliyun OSS (fixed, client open-sourced — but the data was already sent). Policy still permits training on individual plans and China processing. Details in our ZCode agent review.
Credits cost 1× Mon–Fri 14:00–18:00 UTC+8, 0.5× otherwise. Good news for US evenings — but a Chinese workday schedule throttles Asia-based teams exactly when they work.
Since 30.07.2026 the plan is credits-based and legacy V1/V2 plans are no longer sold to new users — conditions have changed before and can change again. Subscriptions are non-refundable.
Pro $80 / Max $168 and the 3-mo/annual discounts do not appear in static docs — confirm live prices before budgeting.
Know the walls before you pay: read the traps above, then decide with eyes open.
See Z.ai (GLM Coding Plan) plansAffiliate link where available — never affects the score.
- official GLM Coding Plan docs — credits, multipliers, peak hours (official) verified
- official DevPack FAQ — Quota Reset Cards expire, non-refundable (official) verified
- official Usage revision notice — credits system since 30.07.2026 (official) verified
- official z.ai subscribe page (JS-rendered — Pro/Max prices unconfirmed in static HTML) unverified
- community HN: Pro weekly 60% in one day, 5h = 20% (2026-07) verified
- benchmark ToolScore internal anchor: 30 days on Pro — 9.63B tokens, 473.94K credits, 92.5% cache hit verified
Data as of 2026-11-11. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
The plan everyone already has access to: $10 buys $15 of multi-model credits (Claude, GPT, Gemini, G
Claude Pro / Max / Ultra — the model everyone else gets compared to, behind the most hated 5-hour +
Kimi Code from the $19 tier, a rolling 5-hour limit with NO weekly cap — and an October 2026 rebuild
Z.ai (GLM Coding Plan) plan review
In-depth analysis · last updated 2026-11-11
Tiers and discounts
Lite from $18 (2,000/5h, 10,000/weekly credits), Pro $80 (12,000/60,000 — our scored tier) and Max $168 (28,000/140,000). Discounts exist but are not in the static docs: 3-month billing comes to $192 (= $64/mo, −20%) and annual is advertised around −30%. Credits convert to tokens with published multipliers — GLM-5.3: (uncached×6.9 + cached×1.7 + output×24) ÷ 10,000; at a 90/8/2 cache-heavy agent profile that is ~256 credits per 1M tokens.
Volume math (the part nobody else publishes)
Pro's official quota is 240K credits/mo ⇒ ~1.6B mixed tokens at peak rates, ~3.2B when ~90% of usage lands off-peak (everything outside Mon–Fri 14:00–18:00 UTC+8 costs half). Validated against our own 30-day anchor: 473.94K credits at 92% off-peak + ~3.7B tokens of Flash bonuses ≈ 9.63B total — the formula reproduces the real number. But 473.94K > 240K official quota: the excess is Quota Reset Cards (official, expiring) and ~300M/day Flash bonuses — the 9.63B is a bonus-inflated upper bound, not the product.
Worst value tier
Volume scales linearly with price (40K/240K/560K credits) — no tier is a worse deal per credit on paper. The practical trap is Lite: the $18 tier's 5h window covers ~20% of a workday's needs and recent buyers report disappointment (HN 50219149). Heavy users get the multiplier effect off-peak; Lite buyers get the smallest slice of it.
Hours before the wall (community)
- Pro: "I spent like 60% of the weekly limits... in one day; each 5 hour limit only gave me 20% to spend" (HN 48811296)
- "5 hour limits felt too restrictive then even for light/medium work" (HN 49295570)
- "lower token limits than the corresponding Anthropic Max subscriptions" (HN 48842100) — vs Claude Max
- "was a decent deal in the past I always ran into limiting with it" (HN 49743070)
- Light users on the old $3 Lite never hit limits (HN 45856628) — legacy era
Client lock-in
20+ agents: the widest official list in the table — Claude Code, ZCode, Codex, OpenCode, AutoClaw, OpenClaw, Cline and more, with MCP tools included.
Versus Anthropic and OpenCode
Against Claude Max, Z.ai wins transparency outright (exact credits vs multipliers-only) but loses on per-window volume per one direct comparison (HN 48842100) and on privacy — no contest after the .git incident. Against OpenCode Go ($10), both are transparent, but OpenCode's budgets are in actual dollars while Z.ai's credits only translate through the multiplier table; Z.ai's counter-argument is breadth (20+ agents vs OpenCode + validated CLIs) and the off-peak half-price window. Best value-per-dollar depends on whether your month is bonus-heavy.