Qwen (Alibaba)

Qwen is Alibaba's model family and, alongside DeepSeek, the reason the open-weight frontier now sits largely in China. The models are strong, the context windows are enormous and the prices are low. Qwen is free to use as a chat assistant and available both as downloadable weights and as a hosted API.

What is Qwen?

Qwen is Alibaba Cloud's large language model family, released both as open weights on Hugging Face and as hosted models through Alibaba Cloud Model Studio. The current flagship is Qwen3.8 Max, released 3 August 2026, priced at $2.00 per million input tokens and $6.00 per million output. Below it sit the 3.6-series models: Qwen3.6 Plus with a 1M-token context window and up to 66K output tokens, and Qwen3.6 Max Preview with 262K context. The family spans dozens of variants covering text, vision, audio, code and embedding, most of which are downloadable and fine-tunable. Qwen models have repeatedly overtaken Western open-weight releases on general knowledge and coding benchmarks.

Key Features

• Qwen3.8 Max: The current flagship, at $2.00/$6.00 per million input/output tokens.
• 1M-token context: Qwen3.6 Plus handles million-token inputs at $0.325 per million input tokens, which is remarkably cheap for that context length.
• Open weights: Most of the family is downloadable, with a huge ecosystem of community fine-tunes and quantisations.
• Multilingual strength: Best-in-class Chinese performance alongside strong results across dozens of other languages.
• Generous free quota: New Alibaba Cloud accounts get 1M free tokens per eligible model on the Singapore endpoint, valid 90 days — and because the quota is per model rather than shared, evaluating across the family adds up to tens of millions of free tokens.

Who Should Use Qwen?

Qwen is a strong choice for developers building multilingual products, teams that need very long context at low cost, and anyone self-hosting open weights. The free evaluation quota makes it unusually easy to benchmark against your own workload before committing. As with DeepSeek, the caveat is jurisdiction: the hosted API runs on Alibaba Cloud, so organisations with data-residency constraints should plan on self-hosting the open weights. English-language documentation is also thinner than what US providers offer.

Pricing

The Qwen chat app at chat.qwen.ai is free. On the API through Alibaba Cloud Model Studio, Qwen3.8 Max costs $2.00 per million input tokens and $6.00 per million output. Qwen3.6 Plus starts at $0.325 per million input and $1.95 per million output with a 1M-token context, and Qwen3.6 Max Preview at $1.03/$6.16 with 262K context. New accounts receive a free quota of 1 million tokens for each eligible model on the International (Singapore) endpoint, valid for 90 days after activating Model Studio — one of the most generous evaluation tiers in the industry. Open weights are free to download and run.

Pros and Cons

Bottom Line

Qwen is one of the two or three most important open-weight families in the world, and Alibaba prices it to win share. For multilingual products, long-context workloads and self-hosted deployments it is a first-choice option, and the free evaluation quota removes any excuse not to test it. Western enterprises with data-residency requirements should run the weights themselves rather than call the hosted API.