ToolScore
LLM APIs 6 min read

Z.ai GLM Coding Plan: Credits → Tokens Math, Plus How to Read Other Plans' Quotas

1 credit ≈ 6.8K–11.8K tokens the number that matters here

By Stan K.

Z.ai is the only provider that publishes its credit math — here is every plan (Lite/Pro/Max) converted to weekly and monthly token volumes, plus why Claude's '5x' and Google's tiers tell you almost nothing.

Coding plans don’t sell tokens. They sell units — credits, requests, “usage,” “compute allowance” — and every vendor picks the unit that makes its limit hardest to compare. This is the decoder ring, and the reason our coding-plans ranking shows a “Value” number at all.

The one plan that publishes its math

Z.ai is the only major provider that prints the conversion in its docs. The GLM Coding Plan bills in credits, and each token type has a published multiplier:

Token typeGLM-5.3GLM-5.3-Flash
Uncached input×6.9×2.3
Cached input×1.7×0.56
Output×24×8

Credits = (uncached×6.9 + cached×1.7 + output×24) ÷ 10,000. Now the units become tokens. A typical agent session runs about 90% cached input / 8% fresh input / 2% output (that’s what a real 92.5%-cache-hit month looks like), which works out to:

  • GLM-5.3: ~256 credits per 1M tokens → 1 credit ≈ 3,900 tokens
  • GLM-5.3-Flash: ~85 credits per 1M → 1 credit ≈ 11,800 tokens
  • A 50/50-ish real-world mix lands near 1 credit ≈ 6,800 tokens

The official Pro quota — 12,000 credits per 5 hours, 60,000 per week — is therefore roughly 1.2B–7.5B tokens a month depending on which model, your cache hit rate, and how much of your usage lands off-peak. Notice the spread: the same credits are ~3× bigger on Flash than on the flagship, and up to 2× bigger again off-peak. Any review that quotes a single token number for a plan is hiding this.

Every Z.ai plan, converted to tokens

Official quotas: Lite $18/mo (2,000 credits/5h · 10,000/week), Pro $80/mo (12,000/5h · 60,000/week), Max $168/mo (28,000/5h · 140,000/week). Z.ai publishes its own “Estimated Token Allowance” table (millions of tokens per week, at a 95% cache hit; the range inside each cell is minimum = 100% peak usage, maximum = 100% off-peak). Monthly = weekly × 4.3:

Plan (per week)GLM-5.3 onlyMixed ~60% FlashFlash only
Lite $1848–97M → 206–416M/mo~80–190M → 343–815M/mo146–292M → 626–1.25B/mo
Pro $80290–580M → 1.24–2.49B/mo~480–1,080M → 2.2–4.5B/mo877–1,755M → 3.76–7.53B/mo
Max $168676–1,352M → 2.9–5.8B/mo~1.1–2.5B → 4.8–9.5B/mo2,047–4,095M → 8.8–17.6B/mo

(Official per-model weekly ranges; the mixed column is our interpolation of a real 60/40 Flash/5.3 agent profile. At 98% cache every number grows another ~8%.)

Official Z.ai quota — credits per 5-hour window
Lite — $18/mo 2,000
Pro — $80/mo 12,000
Max — $168/mo 28,000

Our own 30-day Pro anchor — 9.63B tokens, 63% Flash, 92.5% cache hit, 92% off-peak — sits at the top of the Pro mixed/Flash band exactly as the math predicts, before bonus credits (Quota Reset Cards, daily Flash promos) add anything. So no, the 9.63B month was not “mostly bonuses”: ~4–7.5B of it is what the official quota genuinely delivers to a US-evening, cache-heavy agent user; bonuses covered the rest, and bonuses can vanish.

Orientation only — your tokens will differ

These are quota ranges from the official table, not promises. What you actually get depends entirely on how you work: cache-hit rate (90%+ in long agent sessions, near zero in short chats), output share (output costs 24× more per token than cached input), which model you burn, and your timezone versus the peak window (Mon–Fri 14:00–18:00 UTC+8 — a US workday runs almost entirely at the half-price rate; an Asian one pays full rate at peak hours). Bonus credits can push a real month past every number above — but they expire; the table is the product.

Why Claude’s “5x” tells you almost nothing

Anthropic publishes relative multipliers — Pro is 1×, Max 5x is “five times as much every five hours and weekly,” Ultra 10× weekly with no 5-hour window — but no absolute numbers. You cannot convert “5×” into tokens without a baseline, and the baseline is not published. What the community has measured instead (all receipts in our ranking): a Pro 5-hour window dies in one to three real agent tasks; a Max weekly cap lasts heavy users two to three days. That’s the honest way to compare an opaque plan — user reports, ranges, labeled as such.

Google is further along the opacity curve: Antigravity “Agent requests” come in tiers literally named Limited / Expanded / Higher / Highest, and the community-documented failure mode is nastier than a 5-hour reset — a 654-post forum thread documents lockouts lasting up to 10 days.

The three games to check before you buy

1. The peak-hour window. Z.ai charges full credits 14:00–18:00 Singapore time, weekdays; everything else is 50% off. For a US developer that window is roughly midnight–4am — the entire workday runs at half price, quietly doubling the quota. Buy the same plan in Tokyo and your prime hours pay full rate. No other major plan has this lever.

2. Weekly caps vs 5-hour windows. A 5-hour window refills constantly; a weekly cap does not. Kimi’s new plans advertise “rolling 5-hour limit, no weekly limit” — the friendliest structure on paper. Anthropic’s Ultra is the inverse pitch: no 5-hour limit, but the weekly ceiling is the wall, and heavy users report hitting it inside two days.

3. The bonus fog. Reset cards, launch promos, daily free Flash tokens — all real, all temporary, none with published end dates. Our measured 9.63B-token month ran nearly 2× the official quota thanks to bonuses that can vanish any month. Score the quota, not the promotion.

The rule we use

Published multipliers → we compute tokens and show the math. No multipliers but published prices (OpenCode Go budgets in literal dollars per model) → we convert at the model’s API rate. Neither → “opaque,” and only community-measured ranges, labeled as user reports. No number is invented.

OpenCode’s trick: budgets in dollars

The cleanest idea in the market right now is OpenCode Go’s: your monthly limit is a dollar amount per model (GLM-5.3: $15/mo, GLM-5.3-Flash: $60/mo, and so on), with 5-hour and weekly windows defined as 20% and 50% of it. No credits, no multipliers — just “you get $60 of Flash tokens for $10.” That’s why it tops our value column: you can verify it against any API price list yourself, and the break-even math becomes trivial.

Numbers change — the math doesn't
Token prices and plan limits shift monthly; the structural arguments in this guide age slowly. For current per-tool numbers see the live rankings.