pricingalibabaqwennew-models

Qwen4-Max: $1.50 / $4.50 per Million Tokens, Under Qwen3.8-Max, With Cached Input 3.3× Cheaper

Short answer. Qwen4-Max, released on November 16, 2026, costs $1.50 per million input tokens and $4.50 per million output tokens in Alibaba’s Singapore region, with cache reads at $0.075. Qwen3.8-Max costs 1.3× more on input and output and 3.3× more on cached input. A month of 1,000,000 requests of 2,000 in and 500 out costs $5,250.00.

In plain words. Alibaba’s newest large Qwen model costs less per token than the one before it. Text it has already seen recently costs much less again. A token is a short piece of text, about three-quarters of a word. You pay for the tokens it reads and writes.

What Alibaba released

Alibaba Cloud Model Studio’s model releases page lists, in its Singapore section, “2026-11-16 qwen4-max Service deployment scope: International”. The model page, “Last Updated:Nov 16, 2026”, opens with Alibaba’s own description: “Qwen4-Max is the next-generation flagship multimodal model in the Qwen family, combining text generation, image and video understanding, and deep reasoning.” It gives the model id as “Model invocation ID ( model parameter): qwen4-max”.

The model page prints no price. It says “For prices and free quotas, see Model pricing”, so the figures below come from Model Studio’s central pricing page, read on November 16, 2026. That page says of itself: “This document only lists standard prices”. Promotions, if any, are in the console and not here.

The new-models scan that AI Signal runs every morning saw qwen/qwen4-max appear in OpenRouter’s catalog on November 16. OpenRouter is only a signal for us; the prices are Alibaba’s.

The prices, line by line

Alibaba prices each region separately. As for every Qwen row in our table, the price we publish is the Singapore one, which Alibaba labels “Scope: International”; Beijing and the four Global regions are shown next to it. In the Qwen-Max section of the pricing page, Singapore prints “qwen4-max Cache price (per million tokens) | International | Non-Thinking and Thinking modes | 0<Token≤1M | $1.5 | $4.5 | 1 million tokens |”. The last cell is the free quota. Beijing prints ”| 0<Token≤1M | $1.413 | $4.24 |”, and so do Hong Kong, Frankfurt, Virginia and Tokyo.

The cache prices sit in a small table under each region:

RegionInputOutputCache readCache write (5m)Cache write (1h)
Singapore (International)$1.50$4.50$0.075$1.875$3.00
Beijing and the Global regions$1.413$4.24$0.071$1.767$2.827

The Singapore line on the page is “Cache write (5m) | Cache write (1h) | Explicit/implicit cache read | $1.875 | $3 | $0.075 |”. There is one band, “0<Token≤1M”, so a 900,000-token prompt costs the same per token as a short one. The model page also says “Batch inference | Not supported”, so there is no half-price batch list for this model.

Against Qwen3.8-Max

Qwen3.8-Max sits two lines below on the same pricing page, in Singapore, at ”| $2 | $6 | 1 million tokens |”. Its own model page, read the same day, prints “Input(Implicit Cache) | 0.25 | Per 1M tokens” for Singapore. Alibaba does not mark Qwen3.8-Max as deprecated on either page.

Model, SingaporeInputOutputCached input1M requests, 2,000 in + 500 out
Qwen4-Max$1.50$4.50$0.075$5,250.00
Qwen3.8-Max$2.00$6.00$0.25$7,000.00

Qwen3.8-Max costs 1.3× more on input and on output, and 3.3× more on cached input. The same month costs 1.3× more on Qwen3.8-Max. In Beijing and the Global regions the gap is smaller: Qwen3.8-Max costs 1.2× more on input and output there, and 2.9× more on cached input.

The Qwen team’s announcement says the same in its own words: “We have also lowered API prices compared with Qwen3.8-Max, with particularly large reductions in cached-input pricing.”

Against Gemini 4 Hydrogen, GPT-6.1 Sol and Claude Sonnet 5.5

These are the closest rows we hold from the other large makers, with the date each price was last verified: Gemini 4 Hydrogen on November 13, GPT-6.1 Sol and Claude Sonnet 5.5 on November 9, 2026.

ModelInputOutputCached input1M requests, 2,000 in + 500 out
Qwen4-Max$1.50$4.50$0.075$5,250.00
Gemini 4 Hydrogen$1.50$7.50$0.075$6,750.00
GPT-6.1 Sol$2.00$10.00$0.10$9,000.00
Claude Sonnet 5.5$2.00$10.00$0.10$9,000.00

Gemini 4 Hydrogen has the same input and cached-input prices as Qwen4-Max, and its output costs 1.7× more; the month costs 1.3× more on Hydrogen. GPT-6.1 Sol and Claude Sonnet 5.5 cost 1.3× more on input and cached input and 2.2× more on output; the month costs 1.7× more on either. GPT-6.1 Sol’s price rises for prompts over 272K tokens; the requests here are far below that. The Hydrogen post has Google’s side of that comparison.

Per-token prices are one part of a bill. Qwen4-Max thinks by default, and thinking tokens are output tokens. The model page says: “When reasoning_effort and thinking_budget are omitted, the defaults are reasoning_effort=high and thinking_budget=131072”. How many output tokens a task takes is something we have not measured.

What Alibaba says about task cost

These are Alibaba’s claims, not our measurements. The Qwen team writes: “In our Work360 evaluation, the average cost per task drops from $4.43 to $1.67 — a 62% reduction.” It also writes that “on Work360, token usage drops by approximately 24% and 35% at Xhigh and High effort, respectively”. Work360 is Alibaba’s own evaluation. We have not tested either figure.

Context and output length

The model page gives “Context length | 1,000,000” and “Maximum output length (non-thinking) | 131,072”, plus “Maximum combined tokens for thinking and the final answer | 262,144”. OpenRouter’s page says “It supports up to 393,216 completion tokens.” That is OpenRouter’s figure, not Alibaba’s; our row uses Alibaba’s 131,072.

Where it is available

The Qwen team writes: “Qwen4-Max is now available on QwenCloud. You can start building with either the Token Plan or the API.” On Model Studio the model page lists the Singapore region with “Scope: International” and the Global regions with “Scope: Global”.

There is a second, separate id. The releases page lists qwen4-max-preview on November 11 under “Token Plan”, and its page says “Qwen4-Max-Preview is currently in preview and is available through Token Plan only. Pay-as-you-go access is not supported.” The preview “will be retired after the preview ends”, on November 25. It has no per-token price, so it has no row in our table.

OpenRouter’s price

OpenRouter’s Qwen4 Max page, read on November 16, shows “$1.50 / $4.50 per 1M” and “Released Nov 16, 2026”, served by one provider, “Alibaba Cloud Int.”. Its FAQ gives the same cache figures as Alibaba: “Cache Read at $0.075/M tokens, Cache Creation (5min) at $1.875/M tokens, Cache Creation (1h) at $3.00/M tokens”. It also prints a weighted average actually paid of “$0.2762 /M tokens” for input, which counts cache hits; that is not a list price.

How we checked

On November 16, 2026 we read Alibaba Cloud Model Studio’s pricing page, the Qwen4-Max and Qwen3.8-Max model pages, the model releases page and the Qwen4-Max-Preview page by script, the Qwen team’s announcement in a headless browser, and OpenRouter’s model page and endpoints. Our new Qwen4-Max row matches the Singapore figures, with Beijing and the Global regions stored next to them. The Qwen3.8-Max, Hydrogen, GPT-6.1 Sol and Sonnet 5.5 figures are our dataset rows. Monthly bills and ratios are our own arithmetic in US dollars, without caching or batch, larger price divided by smaller.

Frequently asked questions

How much does Qwen4-Max cost?

$1.50 per million input tokens and $4.50 per million output tokens in Alibaba's Singapore region, with cache reads at $0.075, on Alibaba's Model Studio pricing page read on November 16, 2026. Beijing and the Global regions print '$1.413 | $4.24' for input and output.

Is Qwen4-Max cheaper than Qwen3.8-Max?

Yes: Qwen3.8-Max costs $2 input and $6 output in Singapore, 1.3× Qwen4-Max's prices, and its cached input, $0.25, costs 3.3× Qwen4-Max's $0.075. The Qwen team writes that it has 'lowered API prices compared with Qwen3.8-Max'.

Can I use Qwen4-Max-Preview with pay-as-you-go?

No: its page says 'Pay-as-you-go access is not supported.' and that it is available 'through Token Plan only' until November 25, 2026. Qwen4-Max itself is billed per token.

Drafted with AI assistance from the cited source and reviewed by Álvaro Lucero Rufino, who is responsible for what stays published here. Automated publishing was discontinued in September 2026; posts that no longer met that standard were removed rather than left up. This is original explanatory writing, not a reproduction — read the source above for the full story.