Z.ai Rate Limits by Tier
What are Z.ai's API rate limits?
Z.ai's entry tier (Default) allows an unpublished number of requests/min and an unpublished number of tokens/min for documented default/model family. Limits scale up through Default as cumulative spend and account age increase — see the full table below, verified 2026-08-14.
Limits by tier
| Tier | Qualification | Model class | RPM | TPM | RPD | Concurrent |
|---|---|---|---|---|---|---|
| Default | Account quota shown in the Z.ai console | documented default/model family | — | — | — | — |
— means not documented by Z.ai, never a guess.
What this means for your workload
Classification at volume: 116 calls/min and 60,320 tokens/min at the production profile.
This provider publishes no numeric cap for this workload; check the account console before launch.
Response headers
retry-after | Seconds to wait before retrying, when supplied with a 429 |
rate-limit response headers | Provider-specific remaining and reset counters when documented |
When you exceed the limit
Z.ai returns HTTP 429.
Back off and check the model-specific account quota.
FAQ
What happens when I exceed Z.ai's rate limit?
Z.ai returns HTTP 429. Back off and check the model-specific account quota.
How do I request a rate limit increase on Z.ai?
Request one from the account dashboard: https://bigmodel.cn/console
Decision and evidence guide. Verified 2026-08-14. These are dated, route-local references; unavailable values are not inferred.
Z.ai observed-limit snapshot ledger
Frozen fixture board. Formula / decision rule: effective value = exact account + key + endpoint + model + plan + window observation Boundary: A sibling key, alias, or plan cannot supply an undocumented value.
| Frozen fixture | Joined inputs and observation | Calculated result | State |
|---|---|---|---|
| general API key · exact GLM alias | account=acct-zai-12; key=fp-zai-g-01; endpoint=api.z.ai; model=GLM-5; plan=general; dimension=RPM; window=minute; console=observed 60 The console value is joined to the account, key class, endpoint, exact alias, and minute window. | RPM=60; sharing edge=account-level only where console states it | PASS WITH SCOPE — observed limit. |
| second key under one account · async job | account=acct-zai-12; key=fp-zai-g-02; endpoint=api.z.ai; job=async-14; dimension=concurrency; value=Unavailable; source=response headers The second key joins the account, but no async concurrency ceiling is returned. | concurrency = Unavailable; no copy from general RPM | UNAVAILABLE — live account limit required. |
| coding-compatible endpoint · enterprise override | account=acct-zai-99; endpoint=code.z.ai; model=GLM-Coder; plan=enterprise; override=undocumented; observed-at=2026-08-14 An operator reports an override, but no console or response evidence joins it. | enterprise value = Unavailable; report only the observation provenance | FAIL CLOSED — undocumented override. |
Provenance: zai module 1 first-party evidence and surface verification date 2026-08-14. Z.ai developer limits documentation. Missing joins fail closed.
Z.ai 429 evidence-and-replay classifier
Frozen fixture board. Formula / decision rule: replay = provider evidence ∧ debit visibility ∧ idempotency/effect key Boundary: Backoff changes timing; it cannot make a non-idempotent replay safe.
| Frozen fixture | Joined inputs and observation | Calculated result | State |
|---|---|---|---|
| request-frequency / token-budget 429 | HTTP=429; code=rate_limit; request=req-zai-21; tool=none; reset=5s; debit=visible; idempotency=key-z-1 Returned timing and debit evidence identify a transient request/token class. | retry after 5s with bounded jitter; replay=eligible | RETRY ELIGIBLE — identity joined. |
| partial stream · tool side effect | HTTP=429; request=req-zai-22; stream=partial; tool=charge-card; tool_id=tool-88; effect checkpoint=missing A side-effecting tool call lacks a checkpoint, so replay could duplicate the effect. | retry=blocked regardless of exponential backoff | FAIL CLOSED — operator reconciliation. |
| opaque 429 · insufficient resource or overload | HTTP=429; code=opaque; request=req-zai-23; reset=missing; debit=missing; concurrency=joined The response cannot distinguish resource entitlement from transient overload. | class={resource, overload}; next action=collect console/support evidence | UNRESOLVED — no automatic replay. |
Provenance: zai module 2 first-party evidence and surface verification date 2026-08-14. Z.ai developer limits documentation. Missing joins fail closed.
GLM workload admission board
Frozen fixture board. Formula / decision rule: capacity verdict requires a live account limit for every applicable request/token/concurrency bucket Boundary: Unknown capacity is neither zero nor unlimited.
| Frozen fixture | Joined inputs and observation | Calculated result | State |
|---|---|---|---|
| 100 short calls | account=acct-zai-12; key=fp-zai-g-01; request reserve=100; token reserve=20000; RPM=60; TPM=Unavailable Request bucket alone admits only the first 60 calls; token bucket is unknown. | admitted=60; deferred=40; unknown token impact | DEFER — token ceiling missing. |
| ten 32K prompts · four long outputs · 20 parallel tools | prompt=320000; output=64000; tools=20; concurrency=Unavailable; model=GLM-5 Exact model joins, but concurrency and token reserves do not. | admitted=0; deferred=34; binding bucket=Unavailable | UNAVAILABLE — live limits required. |
| asynchronous batch · retry wave | job=async-19; retry_count=3; shared_pool=account; idempotency=partial; reset=Unavailable The account pool is known, but reset timing and all retry identities are not. | next-safe schedule = Unavailable; do not enqueue retry wave | FAIL CLOSED — incomplete replay joins. |
Provenance: zai module 3 first-party evidence and surface verification date 2026-08-14. Z.ai developer limits documentation. Missing joins fail closed.