IonBin
All pages
Search
Model · Alibaba (Qwen team)

Qwen models (Alibaba)

Qwen is a model family from Alibaba's Qwen team. Some models are released with open weights on Hugging Face, such as Qwen3.8-27B under Apache 2.0. Hosted models, including Qwen3.8-Max and Qwen3.8-Flash, are served through Qwen Cloud, which the Qwen team calls its official API. Review data terms before business use.

Verified Oct 18, 2026 Re-verify: past review date 6 sources

What it is

Qwen is a family of models from Alibaba’s Qwen team. Some models are released with open weights on Hugging Face, and the team serves hosted models through Qwen Cloud, which its model cards call the official Qwen API service. The lineup covers general chat and coding, vision, image generation and speech.

How it works

  1. Hosted. Call Qwen3.8-Max for hard tasks or Qwen3.8-Flash for speed and cost through Qwen Cloud’s API.

  2. Self-hosted. Download an open-weight model such as Qwen3.8-27B and serve it with vLLM, SGLang or Transformers.

  3. The Qwen3.8-27B model card documents thinking enabled by default with a control to disable it. Check the exact hosted model for its supported controls.

Specs and pricing (checked 2026-10-18). Qwen Cloud lists Qwen3.8-Flash at $0.15 per million input tokens and $0.47 per million output tokens, with a 1M-token context. Qwen3.8-27B is released under Apache 2.0 with 262,144 tokens of native context. Larger open releases list other licenses, so read each model card.

Benefits

  • Open weights under Apache 2.0 for some models make self-hosting and fine-tuning possible (see open-weight models).

  • Low hosted prices on the Flash model suit high-volume work on non-sensitive data.

  • Multimodal input on several models covers images and video.

Limits and risks

  • Data terms. Qwen Cloud’s website policy names a Singapore entity, while the customer agreement identifies service entities by jurisdiction. The privacy policy lists user content among data used to maintain and improve services. Distinguish API terms from the separate Try AI provisions before deciding on data use. Get legal or client sign-off before you send client or personal data. Local inference can avoid sending prompts to the model vendor, provided your serving stack, telemetry and external tools do not transmit them.

  • License per model. Not every Qwen release is Apache 2.0. Check the license on each model card.

  • Fast churn. Hosted aliases and downloadable checkpoints have different release cycles. Pin model IDs in production.

This page describes vendor terms as published. It isn’t legal advice; have counsel review data terms before you send client or personal data.

Checked specs

Snapshot from official sources on Oct 18, 2026. Model lineups and prices change often; this page is re-verified every 30 days.

Hosted modelsQwen3.8-Max (snapshot 0902), Qwen3.8-Flash, Qwen-Image-3.0-Pro [3]
Open-weight modelsQwen3.8-27B (Apache 2.0; text, image and video in); larger Qwen3.8 releases use other licenses, so check each card [1]
Qwen3.8-27B context262,144 tokens natively, extensible to 1M; thinking mode on by default [1]
Qwen3.8-Flash price per 1M tokens (checked 2026-10-18)$0.15 input, $0.47 output, $0.016 implicit-cache input [4]
Qwen3.8-Flash limits1M total context; maximum input 991K without thinking or 983K with thinking; maximum output 131K, subject to the total window [4]
Service provider (privacy policy)Website policy names Intelligent Cloud Computing (Singapore) and includes user content among data used to maintain/improve services. [5]
Service agreementContracting entity varies by jurisdiction; distinguish API terms from separate Try AI provisions. [6]

Best-fit stacks

Stack 08

Dev, web & IT

Design, code, host and ship sites and internal tools, with an AI coding agent doing most of the typing.

Stack 04

Marketing content & social

Research, draft, design and review content with an assistant and the publishing tools you already use.

Agent compatibility

API accessYesOfficial docs ↗
Also offered viaQwen Cloud; open weights on Hugging Face

Levels come from each tool's recorded, sourced compatibility. How the levels work.

Sources

  1. Qwen3.8-27B model card · Qwenofficial accessed Oct 17, 2026
  2. Qwen on Hugging Face · Qwenofficial accessed Oct 17, 2026
  3. Qwen Cloud model marketplace · Qwen Cloudofficial accessed Oct 17, 2026
  4. Qwen3.8-Flash · Qwen Cloudofficial accessed Oct 17, 2026
  5. Qwen Cloud privacy policy · Qwen Cloudofficial accessed Oct 17, 2026
  6. Customer agreement · Qwen Cloudofficial accessed Oct 18, 2026

Practitioner views are restated in our words and labeled as views, not measured results. Product facts come from official sources dated above.

Suggest an edit

IonBin's wiki isn't user-edited. Corrections go to our Directory Curator, who checks them against primary sources before anything changes. Spot something wrong or out of date on “Qwen models (Alibaba)”? Email us with a link to a primary source.

Suggest an edit by email or write to contact@ionbin.com