# ModelScope API-Inference

> modelscope.ai | Inference; 1 record: free tier, limits, identity cost and evidence.

HTML https://fshy.ajam.dev/p/modelscope-2/ | JSON https://fshy.ajam.dev/api/v1/entity/modelscope-2.json | how to read a record: https://fshy.ajam.dev/llms-full.txt

## ModelScope API-Inference

- `inference/modelscope-api-inference` | fails-identity | evidence documented | binding request-rate | category inference API, free
- needs card no, phone no, ID yes, institution no | verified 2026-10-28T04:36:18.792Z, checked 2026-10-29 (live) | stale on 2026-11-29
- url https://www.modelscope.ai/docs/model-service/API-Inference/limits
- free tier: Free API access to selected open-source LLM, multimodal and text-to-image models, described as a non-commercial, non-profit product. Quota and concurrency are adjusted dynamically with platform load. Calls can be redeemed with Magicubes: about 0.5 per call for lightweight models, about 1 for standard and about 2 for flagship. A per-day call count is not stated on the page read | unit: requests | metered on: request-count | read on: 2026-10-28 | source: [1] | note: perpetual-free-allowance
- per-object cap: not applicable
- retention: not applicable
- egress: metered: no
- identity: national-id | detail: the ModelScope account must first bind to an Alibaba Cloud account that has completed real-name verification before API-Inference works normally. Whether a non-mainland identity document completes that verification is unverified
- interface: rest
- limits:
  - request-rate (binding): concurrency per model is limited dynamically by platform load, with normal single-concurrency use as the guiding principle; not for high-concurrency or SLA workloads [1]
  - volume-quota: unverified: the daily call quota is not stated on the page read [1]
- binding note: dynamic per-model concurrency limits and an unpublished daily quota
- tooling: part of the free tier
- composes with: free open-model inference for experiments; the per-call Magicube cost is shown next to the provider in each model's code example
- reliability: older models are removed from API-Inference as new models are released
- evidence:
  - [1] 2026-10-28: API-Inference is offered free of charge to developers | capture: sha256:9b81bf0c672324cc38dfa6450b81b748282eb22b0de9f8d6c2b24215fc465047 | quote: offering them free of charge for developers to experience
  - [1] 2026-10-28: the Alibaba Cloud account bound to ModelScope must be real-name verified | capture: sha256:9b81bf0c672324cc38dfa6450b81b748282eb22b0de9f8d6c2b24215fc465047 | quote: Your ModelScope account must first bind to an Alibaba Cloud account. Additionally, to prevent abuse, the corresponding cloud account must have completed real-name verification before you can use API-Inference normally.
  - [1] 2026-10-28: not suitable for high-concurrency or SLA-bound workloads | capture: sha256:9b81bf0c672324cc38dfa6450b81b748282eb22b0de9f8d6c2b24215fc465047 | quote: Do not use it for online tasks requiring high concurrency and SLA guarantees.
  - [1] 2026-10-28: Magicube cost tiers per call | capture: sha256:9b81bf0c672324cc38dfa6450b81b748282eb22b0de9f8d6c2b24215fc465047 | quote: - Lightweight models: ~0.5 Magicubes per call
  - [1] 2026-10-28: older models are removed over time | capture: sha256:9b81bf0c672324cc38dfa6450b81b748282eb22b0de9f8d6c2b24215fc465047 | quote: As new models are released, older models may gradually be removed from API Inference.
- sources:
  - [1] https://www.modelscope.ai/docs/model-service/API-Inference/limits