Vibitely Daily

Tuesday, 17 November 2026, story 1 of 7

Alibaba launches Qwen4-Max, cheaper per task but behind Kimi K3.1 Pro

The 2.4T-parameter model activates 71B per token and costs $3.27 per Artificial Analysis task. Alibaba plans smaller open Qwen4 models later.

Illustration: A large translucent sphere built from many small cubes with only a small cluster lit from within, resting on one pan of a balance scale against a stack of coin-like discs on the other.

What happened

Alibaba has released Qwen4-Max, a mixture-of-experts model with 2.4 trillion parameters, of which 71 billion are active per token, according to the Qwen team. It takes images natively and has a one-million-token context window. It is available through QwenCloud and in QwenWork, Qoder and QwenStudio. Alibaba's API charges $1.50 per million input tokens and $4.50 per million output tokens; cache hits cost $0.075.

Artificial Analysis scores it 43 on its Intelligence Index, three behind Kimi K3.1 Pro at 46. It puts the cost at $3.27 per task, against $3.68 for Kimi and $10.02 for Qwen3.8-Max.

What it means

The pitch is cost per task, not raw ability. Qwen4-Max trails Kimi K3.1 Pro on the index and on Terminal-Bench 4.0 (48 percent against 67), so teams running coding agents may have little reason to switch on price alone. The source gives no price comparison with Opus 5.5 or GPT-6, and the model trails both on Humanity's Last Exam.

The saving over the predecessor comes from lower token prices and fewer output tokens: about 105,000 per task against 171,000. Token use depends on the workload, so a test on your own tasks will likely say more than the index. Open weights are limited: for Qwen4, only smaller models are planned, with no date given.

What to do: Run Qwen4-Max on your own agent tasks and compare cost per finished task, not token price, before deciding whether to move any workload.

Source

Filed under Models. Published .

"What it means" is analysis, not reporting. Found an error? Write to daily@vibitely.com.