Qwen4-Max Launches at $4.50/M Output, 91% Below GPT-6 Astra

Alibaba has launched Qwen4-Max, its new flagship AI model, at a list price of $1.50 per million input tokens and $4.50 per million output tokens on its Model Studio cloud, as Bloomberg reported on 16 November. OpenAI lists GPT-6 Astra at $10 and $50, so Qwen4-Max’s output price is 91% lower, by our arithmetic from both companies’ pricing pages.

The Qwen4-Max price is also 25% below the $2 and $6 Alibaba lists for Qwen3.8-Max, the model it succeeds. Alibaba’s blog post says the model has a 2.4-trillion-parameter backbone and a one-million-token context window, and that it lowered API prices compared with Qwen3.8-Max.

Business Pill · Tokens Explained

An evergreen Business Engineer explainer on how AI models are metered and priced in tokens. It teaches the general idea only and does not discuss Alibaba or Qwen4-Max.

The key insight: As we read it, Alibaba has put a flagship model it says rivals the US labs on its cloud at a list price of $4.50 per million output tokens, against $50 for GPT-6 Astra, 91% lower by our arithmetic. Its own chart shows Qwen4-Max ahead on some agentic tests and behind on others, so the price gap is the figure that does not rest on Alibaba’s own tests.

What Alibaba Launched

In a blog post dated 15 November, the Qwen team wrote that Qwen4-Max has “2.4T backbone parameters, 71B activated per token, and an additional 328B N-gram Embedding parameters.” The post says the model “is now available on QwenCloud” through either the Token Plan or the API. The model is built on what the post calls the new Qwen4 architecture; unlisted Qwen4 models had surfaced in Alibaba’s Qwen Code tool, as we reported on 31 October.

Alibaba Cloud’s Model Studio documentation lists the model ID qwen4-max with a context length of 1,000,000 tokens and a maximum non-thinking output of 131,072 tokens. It accepts text, image and video input, and is compatible with the OpenAI Chat Completions, Responses and Anthropic Messages APIs. The page lists regions including Singapore, Germany (Frankfurt), US (Virginia), Japan (Tokyo), Hong Kong and Beijing.

The launch follows the preview we covered on 11 November. Alibaba’s preview page still says that model “is available through Token Plan only. Pay-as-you-go access is not supported.” The launched qwen4-max now has a pay-as-you-go list price.

Standard output list prices per million tokens for Qwen4-Max and US flagship models, from Alibaba Cloud Model
Standard output list prices per million tokens for Qwen4-Max and US flagship models, from Alibaba Cloud Model Studio (international), OpenAI, Anthropic and Google pricing pages read on 16 November 2026, before cache or batch discounts. Google lists the Gemini 4 Carbon price through 30 November. The 91% gap is our arithmetic.

The Price, Against US Flagships

Alibaba’s pricing page lists qwen4-max, for international deployment, at $1.5 input and $4.5 output per million tokens for requests up to one million tokens, in both thinking and non-thinking modes. Cache reads cost $0.075 per million tokens, 5% of the input price, by our arithmetic.

Against the US labs’ standard list prices on 16 November, by our arithmetic: Qwen4-Max’s output is 91% below GPT-6 Astra ($50), about 89% below Anthropic’s Claude Fable 5.2 ($40), about 78% below Claude Opus 5.5 ($20), and 55% below both OpenAI’s GPT-6.1 Sol and Google’s Gemini 4 Carbon ($10). Google’s page says Gemini 4 Carbon rises to $20 per million output tokens on 1 December.

Alibaba also gives a cost-per-task figure. On its own Work360 evaluation, the post says, “the average cost per task drops from $4.43 to $1.67 — a 62% reduction” compared with Qwen3.8-Max, helped by lower token usage.

Grouped bar chart: Qwen3.8-Max at $2.00 input and $6.00 output per million tokens versus Qwen4-Max at $1.50 and $4.50
Alibaba Cloud Model Studio standard list prices per million tokens for Qwen3.8-Max and Qwen4-Max, international deployment, read on 16 November 2026. The 25% gap is our arithmetic.

What Alibaba Claims on Benchmarks

Every benchmark figure here is Alibaba’s own, from a chart in its post comparing Qwen4-Max with Qwen3.8-Max, Claude Opus 5.5, GPT-6.1 Sol, GPT-6 Astra and Moonshot’s Kimi K3.1. On DeepSWE v1.1, an agentic coding test, Alibaba puts Qwen4-Max at 75.7, against 75.2 for GPT-6.1 Sol, 74.2 for Opus 5.5 and 74.1 for GPT-6 Astra.

The same chart shows Qwen4-Max behind on other tests. On Terminal-Bench 4.0 it scores 48.5, against 59.6 for Opus 5.5, 59.1 for GPT-6 Astra and 67.2 for Kimi K3.1. Of the 16 panels in the chart, Qwen4-Max has the top score in four, by our count.

On Work360, a benchmark Alibaba built itself, the post says “Qwen4-Max scored 52.69, up from 47.74 for Qwen3.8-Max”. Bloomberg reported that Alibaba claimed performance that rivals Anthropic and OpenAI.

Open Weights

Bloomberg reported that Alibaba “did not make the model open weight” but said it would make smaller versions available soon. The Qwen post says: “Smaller models in the family are coming to open source”. The Model Studio page lists fine-tuning and model deployment as not supported for qwen4-max.

When we checked the Qwen organisation’s most recent models on Hugging Face at about 02:50 UTC on 16 November, the newest entries were Qwen3.8 and Qwen3.6 models, and no Qwen4 model appeared in that listing.

The Structural Read

As we read the pricing pages, the gap is widest at the top of the US range: GPT-6 Astra’s output price is about 11 times Qwen4-Max’s, while GPT-6.1 Sol and Gemini 4 Carbon list at $10, so the gap to them is 55%, by our arithmetic.

As we read the post, Alibaba frames cost per task, not only price per token: it says the Work360 cost per task fell 62%, which combines the 25% lower list price with lower token usage, as Alibaba states it.

Qwen Team, Alibaba, 15 November 2026

“In our Work360 evaluation, the average cost per task drops from $4.43 to $1.67 — a 62% reduction.”

Three Implications

FOR DEVELOPERS Model Studio lists qwen4-max as compatible with the OpenAI Chat Completions, Responses and Anthropic Messages APIs, with a 1,000,000-token context.

FOR US LABS Their standard output prices on 16 November run from $10 (GPT-6.1 Sol, Gemini 4 Carbon) to $50 (GPT-6 Astra), against $4.50 for Qwen4-Max, by our arithmetic.

RELATED Epoch AI’s estimate of Alibaba’s AI revenue is in our piece on the China six at $11 billion.

The Business Engineer Lens

This story maps onto the Business Engineer framework Tokenomics: The Economics of AI.

The framework starts from the token and says: “It is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts.”

As we read it, Qwen4-Max puts the price leg of that list in play: Alibaba lists $4.50 per million output tokens against $10 to $50 at the US labs on 16 November, by our arithmetic. The framework predates the launch and does not discuss Qwen4-Max.

What Is Not Established

The benchmark scores are Alibaba’s and were not run independently by us; several panels cite third-party leaderboards, according to the post’s footnotes. The price comparison uses standard list prices before cache or batch discounts, and the cost of a real task also depends on how many tokens each model uses. We read only Bloomberg’s headline and summary, not its full article.

We did not contact Alibaba, OpenAI, Anthropic or Google for this piece.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

Tokenomics: The Economics of AI

The Business Engineer analysis of the token as the unit of price the AI labs charge.

Read the Map of AI →

The Bottom Line

Alibaba lists Qwen4-Max at $1.50 per million input tokens and $4.50 per million output tokens, 25% below Qwen3.8-Max and 91% below GPT-6 Astra’s output price, by our arithmetic. On Alibaba’s own chart it leads GPT-6 Astra on DeepSWE v1.1 and trails it on Terminal-Bench 4.0.

95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. This piece rests on the Qwen team’s blog post announcing Qwen4-Max, which we read in full, including its benchmark chart, and on Alibaba Cloud’s Model Studio model, preview and pricing pages, read on 16 November 2026. Every benchmark score is Alibaba’s own. US prices come from the standard list prices on OpenAI’s, Anthropic’s and Google’s own pricing pages on the same day.

Bloomberg’s report was read as its headline and summary only. The percentage gaps and the benchmark count are our arithmetic. We did not contact the companies named.

Sources: Qwen Team, Qwen4-Max: Your New Teammate Has Arrived, 15 November 2026 · Alibaba Cloud Model Studio, Qwen4-Max model page · Alibaba Cloud Model Studio, Model inference pricing (read 16 November 2026) · Bloomberg, Alibaba Unveils New Flagship AI Model in Riposte to US Rivals, 16 November 2026 · OpenAI, API pricing (read 16 November 2026) · Anthropic, Pricing (read 16 November 2026) · Google, Gemini API pricing (read 16 November 2026)

Scroll to Top
FourWeekMBA