Alibaba's flagship 2.4T-parameter model with a 1M-token context and native vision, priced from $1.50 per million input tokens.

Qwen4-Max launch — Alibaba's new flagship AI model
1 / 2Overview
What is Qwen4-Max?
Alibaba's flagship AI teammate for work, code, and research.
Qwen4-Max is Alibaba's most capable model, released on November 16, 2026. It is a sparse mixture-of-experts model with 2.4 trillion total parameters and 71 billion active per token, built on a new hybrid architecture that combines sparse attention with Gated DeltaNet layers and a 328-billion-parameter N-gram embedding.
The model accepts text, image, and video input with a 1-million-token context window, and targets professional work: agentic coding, long document analysis, finance, science, and creative production. Alibaba also ships a Live Delivery mode that returns a usable first version early and refines it in later passes.
Qwen4-Max is API-only for now. It runs on QwenCloud via the Token Plan and pay-as-you-go API, and is also listed on OpenRouter. Smaller open-weight Qwen4 models are promised for a later release. Pricing starts at $1.50 per million input tokens and $4.50 per million output tokens in Singapore, with cached input at $0.075, and a 1-million-token free quota valid for 90 days.
Platforms and languages
Qwen4-Max Availability
Platforms
Languages
Capabilities
Qwen4-Max Key Features
1M-token context
A one-million-token window for whole codebases, long documents, and multi-step agent runs without losing track of the task.
Native multimodal input
Accepts text, images, and video in a single request, with vision built into the model rather than bolted on.
Hybrid sparse architecture
Three Gated DeltaNet layers compress history before a sparse attention layer picks out what matters, cutting compute on long inputs.
Agentic coding
Built for autonomous coding agents: Alibaba reports 75.7 on DeepSWE v1.1 in its own launch figures, and 46.7 on Agents' Last Exam v1.1.
Live Delivery
Returns a usable first draft early and keeps improving it, so users can steer long-running agent work instead of waiting for a final answer.
Lower cost per task
Alibaba reports about 60% lower average cost per task than Qwen3.8-Max on its own Work360 benchmark, with 25% lower token prices.
Plans and access
Qwen4-Max Pricing
Paid
Common questions
Qwen4-Max FAQs
What is Qwen4-Max?
Qwen4-Max is Alibaba's flagship AI model, released November 16, 2026. It is a 2.4-trillion-parameter sparse mixture-of-experts model with a 1-million-token context window and native text, image, and video input.
How much does Qwen4-Max cost?
On QwenCloud in Singapore it costs $1.50 per million input tokens and $4.50 per million output tokens, with cached input at $0.075. New users get a 1-million-token free quota valid for 90 days.
Is Qwen4-Max open source?
No. The flagship is API-only and its weights are not public. Alibaba says smaller open-weight Qwen4 models will be released later.
Where can I use Qwen4-Max?
Through the QwenCloud API and Token Plan, in the Qwen apps QwenWork and QwenStudio, and via OpenRouter as qwen/qwen4-max.
Reviews
See what the community thinks and share your experience.
—
Based on 0 ratings
Rating distribution
Community reviews
No reviews
No written reviews yet.