Qwen API (Alibaba)
Alibaba's multimodal LLM series — text, vision and code
Freemium
✓ Verified
🇨🇳 China
Alibaba’s Qwen (通义千问) model family, available via Alibaba Cloud DashScope, covers text generation, vision understanding, code generation, and multilingual tasks across 29+ languages. The Qwen3 series includes mixture-of-experts (MoE) models that deliver strong performance at lower cost. Qwen models are the top performers for Chinese language tasks and are widely used in Chinese enterprises. The open-source weights are also available on Hugging Face for self-hosting, making Qwen one of the most versatile Chinese AI offerings for international developers.
API Details
Auth Method
API Key
Pricing Model
Freemium
Free Tier
Yes — limited free tokens monthly
Rate Limit
60 RPM (free)
Format
REST / JSON (OpenAI-compatible)
Versioning
Model-based (qwen-max, qwen-plus, qwen-turbo)
SLA / Uptime
99.5%
Compliance
Alibaba Cloud compliance; Chinese data regulations
Geographic Restrictions
Global; best performance for Chinese language tasks
Last Verified
2026-02-20
Categories
Frequently Asked Questions
Qwen is Alibaba Cloud's large language model series, accessible via the Alibaba Cloud Dashscope API. Qwen2.5 is the latest generation, available in 0.5B to 72B parameter sizes, with instruction-tuned, coding (Coder), and math-specialised variants. It is one of the strongest open-source model families available.
Alibaba Cloud offers a free trial with limited API credits for new users. After that, Qwen pricing is competitive u2014 Qwen-Plus costs around $0.0008 per 1,000 tokens. Pricing varies by model tier (Turbo, Plus, Max). The open-source models can also be self-hosted for free via Hugging Face or ModelScope.
Yes. Despite being developed by a Chinese company, Qwen2.5 is trained on multilingual data and performs well across English, Chinese, Japanese, Korean, and many European languages. It is one of the strongest bilingual (English/Chinese) models available, making it ideal for applications serving both markets.
Yes. Qwen2.5-Coder is a specialised coding variant that performs extremely well on coding benchmarks, competitive with GPT-4o on HumanEval and similar tests. It supports 92 programming languages and is available in 1.5B to 72B parameter sizes. The 32B instruction version is particularly strong for agentic coding tasks.