AI Models. Qwen (Alibaba Cloud)
Generalist / multilingual / code / self-host

Qwen (Alibaba Cloud)

Qwen3.5/3.6 open + Qwen3.7-Max agent-first

Qwen3.5 and Qwen3.6 (open-weight, Apache 2.0) are the current 'workhorse' series in the open ecosystem: various MoE scales, VLM variants (Qwen-VL). Qwen3.7-Max (05/20/2026) is a closed agent-first model via DashScope API. On-prem / cloud / edge; governance and security are the deployer's responsibility.

Verified: 2026-05-22

Purchase decision (when to choose / when to avoid)

Choose if...

  • You want an open-weight 'workhorse' (various scales) and have deployment/self-host competence.
  • You're doing multilingual work and want an alternative outside top-3 Western vendors.
  • Priority is cost and flexibility, not the 'biggest' enterprise brand.

Avoid if...

  • You have sensitive geopolitical/export considerations or a regulated sector — check policies and licenses.
  • You need a wide EU integration ecosystem and enterprise support SLA.

Cost in practice (scenarios)

API in DashScope

Pay-as-you-go; usually competitively priced (check pricing).

  • moderate volume
Self-host

Cost is GPU + maintenance; pays off at steady volume and data control.

  • MLOps
These are estimates/scenarios (not an invoice). Actual cost depends on context length, number of users, limits and retention policies.

Deployment / data / enterprise

Deployment channels

  • DashScope / Alibaba Cloud Model Studio
  • Self-host selected open-weight variants
  • External providers (HF/ModelScope)

Data policy

Training on data
Self-host: on your side; cloud: depends on terms.
Retention
Depends on service.
Data residency
Depends on region/hosting.
Key: license per variant and sector considerations.

Enterprise readiness

Admin
Alibaba Cloud tooling (depends on account and region).
SSO/SCIM
Depends on offering.
Audit
Depends on offering.
DPA
Depends on agreement.
Certifications
Depends on agreement.
Good choice when priority is cost and flexibility of open models.

Best use cases

  • multilingual projects (including Asian languages) and localizations
  • code generation and analysis plus documentation; self-host and fine-tuning (Qwen3.5/3.6)
  • agent pipelines via Qwen3.7-Max (closed API, agent-first).

Strengths

  • Very good open ecosystem (repos, HF); Qwen3.7-Max — agent-first in the cloud.
  • Various MoE scales and variants (text + VLM); good quality in Asian languages and decent in EN/PL.
  • Access via Alibaba Cloud (DashScope) and external providers; low-medium cost.

Weaknesses / risks

  • Licenses depend on variant — watch for exceptions; Qwen3.7-Max API-only (no weights).
  • Smaller ecosystem of ready integrations in Europe than OpenAI/Claude.

Current models (examples)

  • Qwen3.7-Max (05/20/2026) — agent-first, closed API (DashScope).
  • Qwen3.5 / Qwen3.6 — open-weight, various MoE scales; VLM variants (Qwen-VL).

Alternatives (if this model doesn't fit)

MI
Mistral
EU-friendly and often simpler deployments
DE
DeepSeek
cheap reasoning/code at high volume
LL
Llama
larger hosting and tool ecosystem