Skip to main content
The Quantum Dispatch
Back to Home
Cover illustration for Qwen 4 Max Explained: 2.4T Parameters, $1.50 API Pricing

Qwen 4 Max Explained: 2.4T Parameters, $1.50 API Pricing

Alibaba's Qwen 4 Max pairs 2.4 trillion parameters with a 1M-token context at $1.50 per million input tokens. Here is what developers get today.

Dr. Nova Chen
Dr. Nova Chen★Nov 20, 2026★4 min read

Alibaba released Qwen 4 Max on November 16, 2026, and the short version is this: it is the largest and most capable model the Qwen family has shipped, it costs 25% less than the flagship it replaces, and it is the first Max model built on the new Qwen 4 architecture. For developers who have followed Qwen as an open-weight workhorse, there is one more detail worth knowing up front. Qwen 4 Max itself is available through the API and Alibaba's apps, while the open-weight Qwen 4 models are still on the way.

  • Scale: 2.4 trillion total parameters, with 71 billion active per token, according to Alibaba
  • Context and inputs: a 1M-token context window that accepts text, images and video, with text output
  • Price: $1.50 per million input tokens and $4.50 per million output tokens on QwenCloud, with cached input at $0.075
  • Availability: live now through QwenCloud's Token Plan and API, plus QwenWork and QwenStudio

What Is New in the Qwen 4 Max Architecture?

Qwen 4 Max is a sparse mixture-of-experts model, a design in which only a slice of the network wakes up for any given token. Alibaba says the Qwen 4 generation upgrades that design across attention, residual connections and embeddings. The practical result is in the numbers: the model keeps the same 2.4 trillion total parameters as its predecessor but activates 71 billion per token, which Alibaba describes as 25% fewer than Qwen3.8-Max.

Fewer active parameters means less compute per token, and that is where the efficiency claims come from. On Work360, Alibaba's own benchmark covering 721 specialized fields, the company reports that Qwen 4 Max beats Qwen3.8-Max overall while using 35% fewer tokens in high-effort reasoning mode and cutting the average cost per task by around 60%. Those are vendor figures on a proprietary test, so treat them as Alibaba's claim until independent results arrive. If you want the background on why this design scales so well, our sparse mixture-of-experts explainer walks through it.

How Qwen 4 Max Ranks on Coding and Agent Leaderboards

The independent picture so far comes from Arena's public leaderboards. Alibaba says Qwen 4 Max ranked 6th on Code Arena, the head-to-head web development ranking. Decrypt reports a debut score of 1,750 points there, 74 above the previous Qwen flagship, and a 14th-place finish among individual models on Agent Arena, which the outlet notes is the highest of any Chinese entry on that board.

Decrypt also points out that the visual capability here is about understanding images and video and turning them into code or text. It is not an image generator. That makes the model a natural fit for tasks such as rebuilding an interface from a screenshot or reading charts inside long documents.

Qwen 4 Max API Pricing and Limits

The QwenCloud model page lists the full rate card for developers:

  • Input: $1.50 per million tokens
  • Output: $4.50 per million tokens
  • Implicit cache hits: $0.075 per million tokens
  • Maximum output: 131K tokens, or 262K in thinking mode
  • Compatibility: OpenAI-style and Anthropic-style API endpoints, plus function calling, structured outputs, batch processing and web search

The previous flagship was priced at $2 and $6, according to Decrypt, so the list price drops by a quarter before the token savings are counted. The cached-input rate is the one to watch for agent workloads, since long-running agents resend the same instructions on every step.

Will Qwen 4 Open Weights Follow?

Alibaba's announcement says smaller, open-weight models from the Qwen 4 series will be released to developers worldwide soon, without sizes or dates. That matters to a large community. Earlier this year Alibaba opened the weights of its previous flagship, which we covered in our look at the Qwen3.8-Max open-weight release, and builders will be watching to see how much of the new architecture reaches downloadable models.

For now, Qwen 4 Max gives teams a lower-priced frontier option with a very large context window, and a preview of what the next wave of open Qwen models will be built on. We will keep tracking the releases in our AI coverage.

Sources: Alibaba Cloud press room — November 16, 2026; QwenCloud Qwen4-Max model page — accessed November 17, 2026; Decrypt — November 16, 2026.

More AI Stories