Qwen4-Max Goes Live With Prices: $1.50 per Million Tokens
In short
- Qwen4-Max, Alibaba’s new flagship, is now live on OpenRouter with real prices.
- $1.50 per 1M input tokens, $4.50 per 1M output tokens, 1M-token context, text, image and video in.
- That is 25% cheaper than Qwen3.8-Max on both sides, and weights are not open.
Last week Qwen4-Max was a preview with no price tag. Now it has one. If you run an app, an agent or a bot on a big model, this is worth five minutes of your time.
What happened
Qwen published the Qwen4-Max post on November 16, and within minutes the model appeared on OpenRouter as qwen/qwen4-max.
The numbers: $1.50 per million input tokens and $4.50 per million output tokens. Cached input costs $0.075 per million. The context window is 1M tokens. It takes text, images and video, returns text, and reasoning is on by default.
The older Qwen3.8-Max 0902 costs $2 and $6. The faster Prime version costs $4 and $12. So the new flagship undercuts both.
Is it the smartest model out there? No. Artificial Analysis gives it 43 on its Intelligence Index, a bit above the median of 41 for similar models. It is also wordy: it burned 127M tokens in their tests against a median of 78.6M. And it is proprietary. No downloadable weights. The open ones will be the smaller Qwen4 models, which are not out yet.
How to use it
- Long documents. A 1M context fits whole contracts, product catalogs or a month of support chats in one request.
- Image and video input. Product photos, screenshots, short clips. One model for all of it.
- Cached prompts. If your system prompt is long and repeats, cache reads at $0.075 make it nearly free.
- A/B test it. Point your OpenRouter model name at qwen/qwen4-max for one day and compare answers and the bill.
Do the math: say your bot uses 10M input and 2M output tokens a month. On Qwen4-Max that is $15 + $9 = $24 (about €21). On Qwen3.8-Max 0902 it is $20 + $12 = $32. On Prime it is $40 + $24 = $64. You save $8 a month against the first and $40 against the second. Over a year, $96 or $480. Prices are in dollars as billed by the provider, euro figures are approximate.
What it means
Chinese labs keep pushing price down while the big US models stay expensive. Anthropic and OpenAI still lead on top-end quality, but most business tasks do not need the top. They need good enough, fast and cheap.
Watch the speed. Developers on Hacker News already complain that the previous Qwen is slow, around 30 tokens per second. If Qwen4-Max is the same, cheap tokens will not save a chatbot that makes customers wait. Test it before you switch anything.
Read also: Qwen4-Max Preview: Alibaba’s 1M-Token Flagship, Until Nov 25 · GPT-6.1 Terra: $1 per Million Tokens, Half the Old Price · Cheapest AI API in 2026: Price per Million Tokens Compared
Ivan’s Take
Worth a test, not a rush. I would run it on background jobs first: summaries, tagging, product descriptions. Anything where nobody waits for the answer. For customer chat I would check the speed before I move a single request.
Sources: Qwen · OpenRouter · Artificial Analysis




If it doesn’t save you time or money, you won’t see it here
- Facts, not hypeEvery price and number is checked against the official source
- Always the real priceYou see what it costs before you click
- I say when it’s weakEven when it’s OpenAI or Google
