This guide uses source checks from Sep 5, 2026. Provider and gateway prices can change; preserve the checked date with every forecast.
Why This Topic Matters Now
The Sep 4 keyword report still contains a small but useful `glm 5 api` signal from the United States. That is not enough to justify a generic GLM article, but it is enough to publish a practical finance control while Z.AI pricing has a live GLM-5.3-Flash promotional row. Temporary model pricing can help a canary, but it can also damage a forecast if the workbook forgets the expiry condition.
This guide is for Tier 1 and Tier 2 engineering and finance reviewers evaluating GLM routes in a Chinese AI API stack. It separates direct Z.AI rows from AIWave gateway rows, stores the source date, marks the promotional row as temporary, and gives a small OpenAI-compatible request pattern for route acceptance. The goal is a budget that still makes sense after the promotion changes.
Source Facts Checked Today
Z.AI pricing checked on Sep 5, 2026 lists model prices in USD per 1M tokens. The latest-model table lists GLM-5.3-Flash promotional pricing at $0.075 input, $0.015 cached input, and $0.25 output. It lists GLM-5.3 and GLM-5.2 at $1.40 input, $0.26 cached input, and $4.40 output. The text-model table lists GLM-5.1 at $1.40 input, $0.26 cached input, and $4.40 output, GLM-5 at $1.00 input, $0.20 cached input, and $3.20 output, and GLM-4.7 at $0.60 input, $0.11 cached input, and $2.20 output.
The same Z.AI page checked on Sep 5, 2026 lists built-in Web Search at $0.01 per use, GLM-Image at $0.015 per image, CogView-4 at $0.01 per image, CogVideoX-3 at $0.20 per video, and agent pricing rows. Those feature rows should not be averaged into text-token prices. Tool and media use need separate ledger columns, especially for agents that can decide to call search or generate assets during a workflow.
AIWave /api/pricing checked on Sep 5, 2026 returned success=true, 63 records, pricing_version a42d372ccf0b5dd13ecf71203521f9d2, default group ratio 3, and VIP group ratio 1. Parsed GLM gateway examples before account-group math included GLM-5.1 at $2.10 input, about $6.60 output, and about $0.680001 cache-hit input per 1M tokens; GLM-5 at $1.55 input, $4.96 output, and about $0.400001 cache-hit input; and GLM-4.7 at $0.93 input, about $3.41 output, and about $0.220000 cache-hit input.
Planning Matrix
A source-dated planning matrix keeps the page useful for engineers and procurement reviewers. It turns a search query into an auditable route decision instead of a loose model preference.
| Worksheet field | Why it matters | Control |
|---|---|---|
| row_owner | Direct and gateway rows have different terms | Use Z.AI or AIWave explicitly |
| source_date | Rows can move between review and launch | Store Sep 5, 2026 with the row |
| promotion_status | Temporary row should not become annual plan | Set recheck date before approval |
| cached_input | Repeated prefixes may have different rates | Measure cache share by template |
| output_cap | Output can dominate agent spend | Set max tokens by task class |
| tool_fee | Search adds per-use cost | Allowlist and cap tool calls |
| gateway_group | Applied multiplier affects bill | Record default or VIP group |
Implementation Pattern
The implementation pattern keeps credentials as placeholders, pins the AIWave base URL, records the model, and leaves room for route-specific controls. Production applications should move credentials into environment or secret storage.
from dataclasses import dataclass
@dataclass
class GlmBudgetRow:
row_owner: str
model: str
source_checked_at: str
promotion_recheck_at: str
input_per_1m: float
cached_input_per_1m: float
output_per_1m: float
glm_53_flash = GlmBudgetRow(
row_owner="zai_direct",
model="GLM-5.3-Flash",
source_checked_at="2026-09-05",
promotion_recheck_at="2026-09-09",
input_per_1m=0.075,
cached_input_per_1m=0.015,
output_per_1m=0.25,
)
print({"api_key": "YOUR_API_KEY_HERE", "budget_row": glm_53_flash})
Separate Direct and Gateway Rows
A direct Z.AI pricing row is evidence for a direct Z.AI account. An AIWave row is evidence for the gateway route, account group, pricing_version, and unified ledger. A buyer may compare both, but the worksheet should keep them side by side rather than merged. If the direct row is GLM-5.3-Flash and the gateway trial uses GLM-5.1 or GLM-5, the model names should make that difference obvious.
Treat the Promotion as a Timed Input
A promotional price can be a useful trial input, but it should carry an expiry flag. Store current promotional row, normal row if visible, checked date, recheck date, owner, and decision that depends on the row. A budget should fail closed when the date passes. That means finance gets a reminder to refresh the source instead of approving a stale GLM-5.3-Flash assumption.
Measure Cached Input Before Forecasting
GLM pricing tables expose cached input, but a discount is only useful if the workload actually repeats stable prefixes. Store prompt template version, stable context length, volatile context length, measured cache share, model route, and source date. When the system prompt or reference pack changes, start a new cache assumption. This keeps a good canary result from becoming an overconfident monthly forecast.
Put Tool Fees in Their Own Column
Z.AI lists Web Search at $0.01 per use. Agent workflows can turn one user request into multiple tool calls, especially when prompts ask for current facts or external research. Put search calls, image calls, video calls, audio calls, and agent rows in separate columns. A text-token estimate that hides those feature rows will not explain a month-end variance.
Set Route-Specific Output Caps
Output caps should match the GLM task class. A short classifier, a coding review, a migration plan, and a search-enabled analyst response should not share the same max token limit. Record the cap beside the route. If quality improves only when output doubles, the forecast should show that decision explicitly instead of treating longer answers as incidental.
Internal Links for GLM Buyers
Readers should move through AIWave Models docs, Chat Completions docs, Pricing, Trust, the GLM migration ledger, and the price-table normalization guide. The route gives GLM searchers a clear next step without repeating older migration copy.
Procurement Review
Procurement should require row owner, source URL, checked date, promotion status, recheck date, token classes, tool fees, account group, route name, and one redacted request receipt. Engineering should attach acceptance results and usage fields. Finance should preserve both direct-provider rows and gateway rows. Security should confirm that the canary did not expand data exposure to unapproved tools.
Final Checklist
A GLM-5.3 Flash budget is ready when direct and gateway rows are separate, promotion expiry is marked, cached-input behavior is measured, output caps are task-specific, tool calls are governed, and the AIWave route snapshot is saved with pricing_version. Recheck the Z.AI page before approval and recheck AIWave route availability before production traffic.