AI Radar

Alibaba releases Qwen4-Max, a 2.4T-parameter MoE model

Alibaba released Qwen4-Max, its flagship Qwen4 model, described as a 2.4T-parameter MoE with a 1M-token context window. It accepts text, image and video input and returns text, with reasoning on by default. Alibaba says it beats Qwen3.8-Max using up to about 50% fewer tokens, and that the model is not open weight.

What to do
Test Qwen4 Max through OpenRouter against your current model on your own tasks, watching token use and cost.
Who should care
Developers choosing a hosted flagship model for coding, long-context or multimodal work.

Why it matters: Qwen4-Max is available via OpenRouter at $1.5/M input and $4.5/M output with a 1M-token context, but it is not open weight, so it cannot be run locally.

Key facts

2.4T parameters, MoE
1M-token context window
OpenRouter price
$1.5/M input, $4.5/M output
Artificial Analysis Intelligence Index
43
About 67% lower cost per task than Qwen3.8 Max (Artificial Analysis)
Trails Kimi K3.1 Pro on the index, at about 11% lower cost
Alibaba says weights will come only for smaller Qwen4 models

Models

Kimi-K3.1-Pro Moonshot AI

Input
–
Output
–
Context
–
Released
–

qwen4-max Qwen

Input
–
Output
–
Context
–
Released
–
BenchmarkScoreRank
LMArena WebDev1749.8 rating#6

Context

Alibaba says it beats Fable 5.1 and Kimi K3.1 Pro on coding benchmarks, while Artificial Analysis ranks it behind Kimi K3.1 Pro overall.

Timeline

Coverage

  1. New model on OpenRouter: Qwen: Qwen4 Maxopenrouter.ai · OpenRouter

Evidence

High evidence: the claim comes from an official source (the company or project itself), or at least two independent sites reported it.

Details written Nov 16, 11:40 UTC from the sources listed above.