Qwen Rate Limits by Tier
What are Qwen's API rate limits?
Qwen's entry tier (Free quota) allows an unpublished number of requests/min and an unpublished number of tokens/min for documented default/model family. Limits scale up through Paid 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 |
|---|---|---|---|---|---|---|
| Free quota | New account quota | documented default/model family | — | — | — | — |
| Paid | Model Studio account quota | documented default/model family | — | — | — | — |
— means not documented by Qwen, 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
Qwen returns HTTP 429.
Check the region, model, and account quota in Model Studio.
FAQ
What happens when I exceed Qwen's rate limit?
Qwen returns HTTP 429. Check the region, model, and account quota in Model Studio.
How do I request a rate limit increase on Qwen?
Request one from the account dashboard: https://bailian.console.aliyun.com
Batch 49 · qwen decision and evidence contributions. Surface verification: 2026-08-14. These are route-local, server-rendered fixtures; unavailable values are not inferred.
Qwen quota-coordinate resolver
Frozen Batch 49 fixture board. Formula / decision rule: effective limit = minimum(all joined documented ceilings) Boundary: An alias, key, region, or host without a joined ceiling remains Unavailable.
| Frozen fixture / field ID | Joined inputs and observation | Calculated result | State |
|---|---|---|---|
batch49-qwen-m1-r1international workspace · general API key | realm=intl; workspace=ws-qw-17; key=fp-qw-intl-a1; endpoint=model-studio-intl; region=ap-southeast; model=qwen-max; window=minute Workspace, credential, endpoint, region, exact model, and request/token dimensions join. | min(request=60, token=120000, concurrency=8) → request/token/concurrency ceilings retained | PASS WITH SCOPE — frozen coordinate only. |
batch49-qwen-m1-r2two keys sharing one workspace · same GLM alias in two regions | workspace=ws-qw-17; keys=fp-qw-a1/fp-qw-b2; alias=glm-5; regions=intl/mainland; sharing=workspace Key sharing joins at workspace level, but the mainland endpoint has no matching realm evidence. | effective = min(joined workspace ceilings); mainland value = Unavailable | UNAVAILABLE — realm/endpoint join missing. |
batch49-qwen-m1-r3batch submission · undocumented alternate host | job=batch-qw-03; host=alternate-host; model=qwen-plus; request window=Unavailable; token window=Unavailable; source date=2026-08-14 The batch identity exists, but neither the host owner nor a published bucket joins the managed API account. | missing host owner ∨ missing bucket → Unavailable | FAIL CLOSED — no quota invented. |
Provenance: Batch 49 qwen module 1 first-party evidence and surface verification date 2026-08-14. Alibaba Cloud Model Studio rate-limit documentation. Missing joins fail closed.
DashScope throttle-evidence decoder
Frozen Batch 49 fixture board. Formula / decision rule: retry = transient class ∧ returned timing evidence ∧ duplicate-safe request Boundary: An unjoined 429 proves throttling only; it does not identify a bucket.
| Frozen fixture / field ID | Joined inputs and observation | Calculated result | State |
|---|---|---|---|
batch49-qwen-m2-r1request-frequency throttle | HTTP=429; provider_code=QPS_LIMIT; request=req-qw-411; reset=4s; debit=none; idempotency=joined The request ID and reset hint join the request-frequency class; no token debit is observed. | retry after 4s + jitter; replay is safe for the idempotent request | RETRY ELIGIBLE — request bucket only. |
batch49-qwen-m2-r2partial stream · token throttle | HTTP=429; code=TOKEN_LIMIT; request=req-qw-412; stream=partial; debit=observed; tool_effect=none Partial output means the client must persist the checkpoint before any retry; token reset is not returned. | retry timing = Unavailable; resume only with checkpoint and fresh headroom | UNAVAILABLE — reset evidence absent. |
batch49-qwen-m2-r3model-capacity or unknown provider error | HTTP=429; code=UNKNOWN; request=req-qw-413; limit/reset=missing; model=qwen-plus; region=joined Region and model join, but the response cannot distinguish model capacity from account quota. | cause ∈ {capacity, quota}; no bucket-specific retry verdict | UNRESOLVED — operator evidence required. |
Provenance: Batch 49 qwen module 2 first-party evidence and surface verification date 2026-08-14. Alibaba Cloud Model Studio rate-limit documentation. Missing joins fail closed.
Region-aware Qwen admission scheduler
Frozen Batch 49 fixture board. Formula / decision rule: admit only when every applicable realm-local bucket has positive reserved headroom Boundary: Failover cannot cross credential, asset, realm, endpoint, or quota scope without evidence.
| Frozen fixture / field ID | Joined inputs and observation | Calculated result | State |
|---|---|---|---|
batch49-qwen-m3-r160 short chats | realm=intl; prompt=60×800; output=60×200; request reserve=60; token reserve=60000; cap=120000/min All reserves fit the joined international request and token buckets. | headroom = (60−60, 120000−60000) = (0, 60000); admit=60 | ADMIT — request bucket binds. |
batch49-qwen-m3-r2eight 64K prompts · four long outputs · 20 tools | prompt=512000; output=128000; tools=20; token cap=Unavailable; concurrency=8 Concurrency is joined, but token capacity for the exact long-context model is not. | admitted=0; deferred=32; unknown=token workloads | UNAVAILABLE — live token ceiling required. |
batch49-qwen-m3-r3cross-realm failover wave | source=realm-intl; target=realm-mainland; key=fp-qw-intl-a1; model=qwen-max; asset=missing Target realm credential and model asset are not joined to the source workload. | cross-realm failover = blocked when any identity join is missing | FAIL CLOSED — retain or requalify locally. |
Provenance: Batch 49 qwen module 3 first-party evidence and surface verification date 2026-08-14. Alibaba Cloud Model Studio rate-limit documentation. Missing joins fail closed.