This guide uses source checks from Sep 1, 2026. Provider and gateway prices can change; preserve the checked date with every forecast.
Why This Topic Matters Now
The Sep 1 publishing cycle starts from a narrow signal in the Aug 31 keyword report: `glm 5 api` appeared in Tier 1 search data, while `glm deepseek kimi` appeared as a new seven-day growth candidate. That is not enough volume for a broad GLM campaign, but it is enough to write a practical page for engineers who are already comparing Chinese model families and need a migration control, not a slogan.
This guide is for Tier 1 and Tier 2 API teams evaluating GLM routes beside DeepSeek, Kimi, and Qwen. It uses live AIWave pricing checked on Sep 1, 2026 and current Z.AI pricing docs checked the same day. The core advice is simple: before moving a GLM workload to a newer route, preserve row owner, checked date, model ID, cache behavior, tool-use policy, output cap, and acceptance result in one ledger.
Source Facts Checked Today
AIWave /api/pricing checked on Sep 1, 2026 returned success=true, 63 records, pricing_version a42d372ccf0b5dd13ecf71203521f9d2, supported OpenAI chat completions metadata, default group ratio 3, and VIP group ratio 1. Parsed live gateway examples before account-group math included GLM-5.1 at $2.10 input, $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-5-Turbo at $1.80 input, $5.40 output, and about $0.480001 cache-hit input.
Z.AI pricing checked on Sep 1, 2026 lists prices in USD. The latest-model table lists GLM-5.3 and GLM-5.2 at $1.40 input, $0.26 cached input, and $4.40 output per 1M tokens. It also lists GLM-5.3-Flash at a promotional $0.075 input, $0.015 cached input, and $0.25 output until 24:00 on September 9, 2026 UTC+8. The text-model table lists GLM-5.1 at $1.40 input, $0.26 cached input, and $4.40 output, and GLM-5 at $1.00 input, $0.20 cached input, and $3.20 output.
The same Z.AI page checked on Sep 1, 2026 lists Web Search at $0.01 per use, image generation rows such as GLM-Image at $0.015 per image, audio rows, video rows, and agent rows. A GLM migration therefore cannot be reviewed as text tokens only if the application enables search, media, or agent features. Keep those feature families in separate ledger columns so finance can explain month-end movement without rereading application prompts.
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.
| Ledger row | Migration risk | Control |
|---|---|---|
| row_owner | Direct Z.AI prices and AIWave gateway rows get blended | Store provider, gateway, and internal measured rows separately |
| checked_date | A promotion or model row expires after the forecast | Attach Sep 1, 2026 source date to every imported number |
| model_id | GLM-5.3 candidate silently replaces GLM-5.1 | Pin route names in configuration |
| cache_class | Cached-input assumptions are not measured | Store cache-hit, cache-miss, or unknown |
| tool_calls | Search calls create extra spend | Cap and log per user-visible request |
| output_cap | Long reasoning output dominates cost | Set route-specific max tokens |
| account_group | Finance cannot reproduce applied multiplier | Record effective group with pricing_version |
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 openai import OpenAI
client = OpenAI(api_key="YOUR_API_KEY_HERE", base_url="https://aiwave.live/v1")
GLM_ROUTE = {
"model": "glm-5.1",
"pricing_version": "a42d372ccf0b5dd13ecf71203521f9d2",
"source_checked_at": "2026-09-01",
"max_tokens": 800,
"web_search_allowed": False,
}
def run_glm_acceptance(prompt: str):
response = client.chat.completions.create(
model=GLM_ROUTE["model"],
messages=[{"role": "user", "content": prompt}],
max_tokens=GLM_ROUTE["max_tokens"],
temperature=0.2,
)
return {
"model": response.model,
"finish_reason": response.choices[0].finish_reason,
"usage": response.usage.model_dump() if response.usage else None,
"pricing_version": GLM_ROUTE["pricing_version"],
}
Start With a Migration Ledger
A GLM migration should begin as a ledger, not as a model-name change. Create one row per acceptance run with source URL, checked date, route owner, model ID, prompt-template version, input tokens, cached-input tokens when visible, output tokens, tool calls, status, and reviewer decision. This turns a model upgrade into something engineering, security, and finance can all audit.
Separate Z.AI Rows From AIWave Rows
Z.AI owns its direct-provider pricing table, model docs, and feature pricing. AIWave owns the gateway route, pricing_version, account group, and OpenAI-compatible endpoint evidence. A buyer may compare both, but the workbook should never collapse both into one blended row. If the direct page lists GLM-5.3 and the gateway trial uses GLM-5.1, the ledger should say exactly that.
Treat Promotions as Date-Bound Inputs
The GLM-5.3-Flash promotional row has an explicit September 9, 2026 UTC+8 end date on the Z.AI pricing page checked today. That makes it useful for short experiments, but risky as the sole basis for an annual forecast. Mark promotional rows as date-bound, include list-row context when visible, and require a recheck before any procurement decision that extends beyond the promotion window.
Measure Cache Behavior Before Forecasting
Cached input deserves its own ledger column because both Z.AI and AIWave expose cache-related pricing dimensions. Do not assume repeated prompts will keep cache-hit behavior after product copy, schemas, policy preambles, or tool instructions change. Store prompt-template version and reference-pack version beside token fields so a later variance review can find the cause.
Cap Tool Calls
Search and agent features can be valuable, but they make a GLM trial harder to govern. If the workload does not need current external facts, disable search for the first run. If it does need search, set a per-request ceiling and log each use beside the model call. The goal is not to suppress useful tools; it is to prevent a single vague prompt from turning into unreviewable tool spend.
Use Internal Links for GLM Searchers
Readers arriving through GLM queries should move through Models docs, Chat Completions, Pricing, Trust, the earlier GLM migration guide, and the success-rate denominator guide. Those links turn model interest into a trial path.
Run a Canary Before Promotion
A GLM route can pass a single demo and still fail a production rollout. Use a canary with redacted prompts, bounded output, fixed tool policy, and a written rollback rule. Keep the previous model route available until the acceptance set passes and the cost ledger matches expectations. Promotion should be a configuration release with an owner, not an untracked string edit.
Procurement Review
Procurement should ask for the AIWave pricing_version, Z.AI source URL, checked date, model row, account group, cache policy, tool policy, and one redacted acceptance result. Engineering should explain which rows are direct-provider facts, which rows are gateway facts, and which rows are measured workload facts. That is enough to approve a controlled trial without implying all GLM routes are equally ready.
Final Checklist
A GLM-5.3 migration plan is ready when source rows are date-bound, model IDs are pinned, cached input is measured, tool calls are capped, output budgets are set, account group is recorded, and the canary has a rollback rule. Recheck both Z.AI and AIWave before launch; stale rows are the easiest way to turn a useful migration into a billing dispute.