Qwen is the one major Chinese model family we had not written a proper access guide for — an odd gap, given Alibaba's models are official-channel supplied on Turiloop and the Qwen3.8 preview has made the family front-page news. This closes the gap: what the shipping Qwen tier actually is, how to call it from anywhere with an OpenAI SDK, and how to choose between Max and Plus.
What you can call today
The generation with a real, generally available API is Qwen3.7, in two tiers:
- `qwen3.7-max` — the flagship tier: Alibaba's strongest generally-available model, strong across general reasoning, writing and multilingual work. On Turiloop it runs at $2.50 per million input tokens (output and cache rates on the model page).
- `qwen3.7-plus` — the balanced tier at $0.40 per million input tokens: the economical default when you want Qwen's behavior without flagship pricing.
Qwen3.8-Max-Preview is not this. It exists, it is loud, and it is currently reachable mainly through Alibaba's own Token Plan subscription bundles rather than a standard API — the full state of play is in our Qwen3.8 tracker. When a proper API lands, it joins the lineup; until then, 3.7 is what production runs on.
The setup: two lines
Qwen through Turiloop speaks the OpenAI interface, so there is no SDK to learn:
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["TURILOOP_API_KEY"],
base_url="https://api.turiloop.com/v1",
)
r = client.chat.completions.create(
model="qwen3.7-max",
messages=[{"role": "user", "content": "Summarize this contract clause in plain English: ..."}],
)
print(r.choices[0].message.content)Streaming, tool calling and system prompts work exactly as they do on any OpenAI-compatible model — switch model to qwen3.7-plus and nothing else changes. That is the entire migration story: international card, pay-as-you-go, no Chinese phone number, and the same key also reaches GLM-5.2, DeepSeek V4 and Kimi.
Max or Plus: a practical split
- Default to `qwen3.7-plus` for classification, extraction, summarization, routine chat — the volume work where $0.40 input pricing compounds.
- Escalate to `qwen3.7-max` for long-form writing, multilingual output, and reasoning where the flagship gap shows.
- Coding is the exception: for code-heavy work the lineup's specialists — GLM-5.2 or DeepSeek V4-Pro — beat both Qwen tiers per dollar; see the coding guide.
The pricing calculator has the full Chinese lineup for running your own monthly math.
FAQ
Can I use the Qwen API without a Chinese phone number? Yes — through an OpenAI-compatible gateway like Turiloop: international card, pay-as-you-go, no mainland account required. Point your SDK's base_url at api.turiloop.com/v1 and set the model to qwen3.7-max or qwen3.7-plus.
What is the difference between Qwen3.7-Max and Plus? Max is the flagship tier ($2.50/M input on Turiloop) for reasoning, writing and multilingual work; Plus is the balanced tier ($0.40/M input) for volume tasks. Same API shape — switching is one string.
Is Qwen3.8 available via API? Not broadly. Qwen3.8-Max-Preview currently ships through Alibaba's Token Plan subscriptions and Qoder tools; a standard API listing has not rolled out. Track the state in our Qwen3.8 page.
Is Qwen good for coding? It is capable, but the per-dollar coding leaders remain GLM-5.2 and DeepSeek V4-Pro. Use Qwen where its general and multilingual strengths matter, and route code-heavy work to the specialists.