Alibaba
Qwen4-Max is generally available at $1.50 per 1M input tokens
Alibaba made Qwen4-Max generally available through its API at $1.50 per 1M input tokens and $4.50 per 1M output, with thinking on by default.

Alibaba has made Qwen4-Max, the most capable model in its Qwen family, generally available through its API at $1.50 per 1M input tokens. Thinking is on by default at high effort, and the weights are closed.
Qwen4-Max Is Generally Available in the API
Alibaba released Qwen4-Max on Nov. 16, according to Alizila, the company’s news site. It is available on QwenCloud through the Token Plan and the API, and in Alibaba’s QwenWork and QwenStudio platforms. Alibaba said smaller open-weight models from the Qwen4 series “will soon be released.”
The Price Is $1.50 Input and $4.50 Output
Alibaba Cloud’s Model Studio documentation, updated Nov. 16, lists the international Singapore deployment at $1.50 per 1M input tokens and $4.50 per 1M output tokens. The rate is the same with thinking on or off, for requests up to 1M input tokens. Regions including US Virginia and Frankfurt list at $1.413 input and $4.24 output.
Alibaba said a cached input hit costs $0.075 per 1M tokens. The international deployment comes with a free quota of 1M tokens, valid for 90 days.
OpenRouter listed Qwen3.8-Max (0902) at $2 input and $6 output per 1M tokens on Nov. 16. Alibaba’s Qwen4-Max price is 25% lower on both counts, though the two figures come from different sellers.
The Model Takes Video and Thinks by Default
Qwen4-Max accepts text, image and video and returns text, with a 1M-token context window. Alibaba said it is a sparse mixture-of-experts model with 2.4 trillion total parameters and 71 billion activated per token, 25% fewer activated than Qwen3.8-Max at the same total size.
Output runs to 131,072 tokens without thinking. With thinking on, the reasoning and the answer together can reach 262,144 tokens. Thinking is on by default, at the “high” reasoning effort, according to the documentation.
Existing Code Can Point at It
The API speaks the OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and DashScope formats. It supports web search, function calling, parallel tool calls and structured output. Batch inference, fine-tuning and model deployment are not supported.
Tier 1 rate limits on the Singapore deployment are 5M tokens per minute.
Alibaba’s Own Benchmarks Claim a 60% Cost Cut
Alibaba said on Nov. 16 that Qwen4-Max ranks 6th on Code Arena, a coding leaderboard. On Work360, Alibaba’s proprietary benchmark covering tasks in 721 specialized fields, Alibaba said Qwen4-Max beats Qwen3.8-Max overall. At high reasoning effort it uses 35% fewer tokens and cuts average cost per task by about 60%, the company said. Both results are Alibaba’s own claims.
Our View
We think the default reasoning setting will shape a bill more than the list price. Thinking runs at high effort unless a developer changes it, and output costs $4.50 per 1M tokens against $1.50 for input. Thinking and the answer together can run to 262,144 output tokens, so a short prompt can cost far more in output than in input.
The token savings come from Alibaba’s own Work360 benchmark, and a team’s tasks may use tokens differently.
Developers paying per token can test that without rewriting anything, because the API accepts OpenAI and Anthropic formats. They can send a sample of real jobs and compare total tokens and cost against their current model.
The saving holds only if Qwen4-Max uses far fewer tokens on a team’s own tasks than the model it replaces.
