Qwen API
The Qwen API in Europe
Qwen's open-weight models are published under permissive licences, so you can call them from European infrastructure rather than from Alibaba Cloud. On Frontière AI, 6 Qwen models are live: 5 of them run at European operators and are labeled EU sovereign — Qwen3 235B (instruct), Qwen3.5 397B (multimodal), Qwen3.6 35B A3B (multimodal), Qwen3 Coder 30B and Qwen3.8 27B (multimodal) — and Qwen3.8 Flash (Qwen) is served by US providers, labeled fast access before you call it. Alibaba Cloud's own Model Studio has a Frankfurt region, but its documentation limits EU-only inference there to three of its API models. Same OpenAI-compatible API, prepaid, prices VAT included.
Qwen models on Frontière AI
Every Qwen model we serve, with the provider that serves it and its jurisdiction label — taken from our live catalog when this page is built.
| Model | Label | Primary provider | Price per 100k tokens (in / out) | Tool calling |
|---|---|---|---|---|
Qwen3 235B (instruct)qwen3-235b | EU sovereign | Scaleway | €0.126 / €0.378 | Yes |
Qwen3.5 397B (multimodal)qwen3.5-397b | EU sovereign | Scaleway | €0.1008 / €0.6048 | Yes |
Qwen3.6 35B A3B (multimodal)qwen3.6-35b | EU sovereign | Scaleway | €0.042 / €0.252 | Yes |
Qwen3 Coder 30Bqwen3-coder-30b | EU sovereign | Scaleway | €0.0336 / €0.1344 | Yes |
Qwen3.8 27B (multimodal)qwen3.8-27b | EU sovereign | Scaleway | €0.1008 / €0.5544 | Yes |
Qwen3.8 Flash (Qwen)qwen3.8-flash | Fast access (US) | OpenRouter | €0.0252 / €0.0773 | Yes |
Prices margin and VAT included — the amount debited from your prepaid balance. Tool calling measured with one real call per model.
Detailed Qwen API pricing and cost calculator →Not callable right now: Qwen3.6 27B (multimodal) (temporarily paused since 2026-10-20) · Qwen3.5 9B (fast, budget) (temporarily paused since 2026-10-20) · Qwen2.5-VL 72B (vision) (temporarily paused since 2026-10-20) · Qwen3.8 Max (2.4T, 1M context) (temporarily paused since 2026-09-23)
Alibaba Cloud's own API, side by side
What Alibaba Cloud's Model Studio documents say, checked on 2026-10-12, next to the same questions on Frontière AI. We quote; we do not interpret on their behalf.
| Question | Alibaba Cloud Model Studio | Frontière AI |
|---|---|---|
| Contracting company for EEA customers | Alibaba (Netherlands) B.V. (membership agreement) | FLEECE AI SASU, France |
| EU region | Germany (Frankfurt); “EU” deployment scope offered for qwen-plus, qwen3.5-flash and qwen3-vl-plus only (regions and pricing pages) | Yes, for the models labeled EU sovereign |
| Open-weight Qwen models | Offered under the “Global” deployment scope; the docs say Global trades geographic boundaries for a larger inference pool | Served at the provider named on each model — European for the sovereign ones |
| Prompt retention | Prompts and responses kept for no longer than 30 days by default; zero data retention on request for enterprise customers on listed models | API prompts and completions never written to our database — only token counts |
| Training on your data | “will never use your data for model training” (privacy notice) | We do not train models; provider policies are documented per model |
Alibaba Cloud sources, checked on 2026-10-12: Alibaba Cloud — Model Studio regions · Alibaba Cloud — Model Studio pricing · Alibaba Cloud — data retention policy · Alibaba Cloud — Model Studio privacy notice · Alibaba Cloud — membership agreement · Hugging Face — Qwen
Why an EU provider can serve Qwen
The Qwen team publishes most of its models' weights on Hugging Face under Apache 2.0, and each model page here shows the licence of the exact weights we serve. Anyone can host them, which is how OVHcloud and Scaleway run Qwen models in France with no request reaching Alibaba Cloud.
Switching takes one line
If your code already uses the OpenAI SDK — Model Studio's compatible mode works the same way — only the base URL, the key and the model ID change:
from openai import OpenAI
client = OpenAI(
base_url="https://getfrontiereai.eu/api/v1",
api_key=os.environ["FRONTIERE_KEY"],
)
reply = client.chat.completions.create(
model="qwen3-235b", # EU sovereign: served in Europe, labeled before you call it
messages=[{"role": "user", "content": "Bonjour"}],
)FAQ
Can I use the Qwen API in Europe?
Yes. Alibaba Cloud's Model Studio has a Frankfurt region, where its documentation offers EU-only inference for three API models (qwen-plus, qwen3.5-flash, qwen3-vl-plus); its other models run under a Global scope. Qwen's open-weight models can also be called from European operators: on Frontière AI, 5 Qwen models are labeled EU sovereign, through an OpenAI-compatible API.
Which Qwen models are hosted in the EU on Frontière AI?
Qwen3 235B (instruct), Qwen3.5 397B (multimodal), Qwen3.6 35B A3B (multimodal), Qwen3 Coder 30B and Qwen3.8 27B (multimodal) — labeled EU sovereign and served by European operators. Qwen3.8 Flash (Qwen) is served by US providers and labeled fast access. Each model page states its provider, licence and measured tool-calling support.
Is Qwen open source?
Most Qwen models are published with open weights under Apache 2.0, which allows commercial use and hosting by third parties; a few recent models carry their own licence. Our model pages show the licence of the exact weights we serve, as listed on Hugging Face.
How much does the Qwen API cost on Frontière AI?
The cheapest on input is Qwen3 Coder 30B: €0.0336 in and €0.1344 out per 100k tokens. The table on this page gives every model's price. Prices include margin and VAT, billed per token against a prepaid balance from €10 — no subscription.
Do I need to change my code?
No more than the base URL, the key and the model ID: set the base URL to https://getfrontiereai.eu/api/v1, use a Frontière AI key, and pick one of our model IDs, such as qwen3-235b.
Call Qwen from Europe
Create an account, top up from €10, and use the same OpenAI-compatible API.
Create an account