Price terms as published

Each line is the vendor’s own wording, split from the price capture checked on 2026-09-08. Nothing here is rounded, converted or re-worded — the full sentences stay in the review below. Lines that carry only a price continue the plan named on the line above, in the vendor’s own order.

2 price terms for ModelScope API-Inference
#Plan / termChecked
1Free: 2,000 calls/day total, ≤200/day per model2026-09-08
2overage = HTTP 4292026-09-08

One-liner: ModelScope (Alibaba’s open-model community) exposes a large slice of its model zoo through an OpenAI-compatible API at api-inference.modelscope.cn/v1, free up to 2,000 calls per user per day.

Verified facts (checked 2026-09-08)

Attribute Value Source
Daily quota 2,000 API-Inference calls/day total, all models shared official free-resources page (MSEO, updated 2026-08-14), read same day
Per-model cap Dynamic, max ≤200/day per model, “subject to platform adjustment” official page, same
Reset Daily reset (no rollover); overage returns 429 official page + Cherry Studio docs cross-check
Model catalog “60,000+ models available” claim on the free-resources page; /v1/models returned 50 rows on our same-day probe (the listing is paginated/truncated, treat 60K as catalog-wide, not API-wide) official page + direct API read
Catalog spot-check deepseek-ai/DeepSeek-V4-Pro-0813 ✅ present, Qwen/Qwen3-235B-A22B ✅, MiniMax/MiniMax-M3 ✅, ZhipuAI/GLM-5.2 ✅; ZhipuAI/GLM-5.3 — listed two days ago in community posts — absent today direct API read 2026-09-08
Auth ModelScope SDK token (personal center → access tokens → write-permission token) used as the API key official docs / community cross-check
Prerequisite Alibaba Cloud account binding required — free inference runs on Aliyun compute official free-resources page
Positioning Official wording: for development, testing, learning, prototype validation — not for high-concurrency/low-latency production with SLA official page §service positioning
Format OpenAI-compatible chat completions; Anthropic-format endpoint documented for Claude Code style clients official docs

The MagicPoints trap (what viral posts get wrong)

Several viral articles claim ModelScope’s “MagicPoints” can be spent on API calls — “300 points = 150 flagship calls.” The official pages separate these systems: MagicPoints (150/day from login + up to ~450/day from tasks) pay for the AIGC image/video/training playground, not API-Inference, which is metered in calls, not points. Two different wallets. Don’t blend them into a single “free tokens” story.

Relatedly: do not compare free tiers across gateways by raw numbers. ModelScope sells 2,000 calls/day; Cline API sells dollar credit + a rotating :free subset; VibeX advertises 10 billion tokens/day; Agnes prices per-million-tokens at ¥0. Calls, dollars, and tokens are not convertible currencies.

What it’s genuinely good for

  • Prototype routing across open models — DeepSeek, Qwen, GLM, MiniMax in one endpoint with one token, no VPN and no card required.
  • Agent experiments where per-model caps don’t bite — 200 calls/day on your primary plus fallbacks spread the limit.
  • Not a production dependency. The official positioning disclaims SLA, and our two-day catalog diff (GLM-5.3 in community lists, gone from the API today) shows exactly the churn the deprecation notice warns about: older models get throttled, then retired.

The pattern

ModelScope is the community-platform funnel: Alibaba’s way of keeping the open-model distribution layer (and its compute billing relationship) under one roof, the way Hugging Face monetizes inference. In the 2026 free-tier retreat — when several Western gateways shrank their giveaways — a flat 2,000-calls/day pool makes the open-model ecosystem’s head tier look unusually loose. The cost is lock-in shape: real-name Aliyun binding is the ticket of admission.

Sources