Alibaba Launches Qwen4-Max, Its New Flagship AI Model

Admin


Alibaba's Qwen team launched the full Qwen4-Max model on November 16, 2026, five days after opening it as a preview limited to its Token Plan subscription. Qwen4-Max is the flagship of the new Qwen4 family and the successor to Qwen3.8-Max, built on a new Mixture-of-Experts architecture with native vision and a 1-million-token context window. Alibaba describes it as a multimodal model for text, image, and video understanding, designed for coding, office productivity, front-end design, and complex agentic tasks, and the company calls it "your new AI teammate for work, creativity, and discovery."

The launch matters most for what it changes about access and price. Unlike the preview, Qwen4-Max now has a public pay-as-you-go API on QwenCloud, with the model ID `qwen4-max`, priced at $1.50 per million input tokens and $4.50 per million output tokens, with implicitly cached input at $0.075 per million. Alibaba says all three prices are lower than Qwen3.8-Max, and claims the new model outperforms its predecessor while using up to about 50% fewer tokens, cutting average cost per task by more than 60% in its internal evaluation. The model is also available in Qwen Studio, QwenWork, and Qoder.

Early independent results suggest a genuine step up, though not a clean sweep. On launch day, Qwen4-Max debuted at number 6 on Code Arena's WebDev board with a preliminary 1,750 points, 74 points above Qwen3.8-Max, and at number 14 overall on Agent Arena, up from number 39 for its predecessor and the top Chinese model on that board. Artificial Analysis scored it 43 on its Intelligence Index at the highest reasoning setting, nine points ahead of Qwen3.8-Max and three behind Kimi K3.1 Pro at 46. The same benchmark, however, has Qwen4-Max at 48% on Terminal-Bench 4.0 against 67% for Kimi K3.1 Pro. Alibaba says open weights of smaller Qwen4 models will follow, without giving a date; Qwen4-Max itself is available only through QwenCloud and Qwen's apps.

Why it matters

The flagship tier of AI models is turning into a price war, and Qwen4-Max is Alibaba's sharpest entry yet. At $1.50 per million input tokens, a top-tier model now costs a fraction of what Western flagships charge, and Alibaba's efficiency claims — fewer tokens per task, lower cost per task — attack the metric that actually decides which model developers put into production. Raw benchmark scores make headlines; cost per completed task decides budgets.

The launch also narrows the gap between Chinese and Western labs on the work that matters most right now: agentic coding and long multi-step tasks, where Qwen4-Max's Arena debuts are strong. The Terminal-Bench gap behind Kimi is a reminder that the field is crowded and no single release settles it. But with a public API, a 1-million-token context, and open weights promised for smaller siblings, Alibaba is making it harder for any developer to ignore the Qwen family when choosing a model.

Source: The Model Atlas — Qwen4-Max is out: price, benchmarks and how to try it (November 16, 2026). Summary rewritten by VCTRIA.

Post a Comment