Alibaba Launches Qwen4-Max, a 2.4 Trillion Parameter Flagship With a 1M Token Context Window and Lower Prices
📢 Meet Qwen4-Max, your new AI teammate for work, creativity, and discovery! Qwen4-Max is built on a new MoE architecture with native vision and a 1M-token context window. Highlights: - Outperforms Qwen3.8-Max with up to ~50% fewer tokens, while average cost per task drops over 60% in our internal evaluation. - Advances real-world productivity through greater proficiency with professional tools, stronger long-horizon execution, and higher-quality deliverables. - Handles real tasks, including 3D modeling, game development, manufacturing, finance, law, healthcare, and frontier scientific discovery. - $1.5/M tokens for input, $4.5/M tokens for output, and $0.075/M tokens for implicit caching — all lower than Qwen3.8-Max, with implicit caching down by 70%. The API is live on QwenCloud. You can also try Qwen4-Max on Qwen Studio, QwenWork and Qoder. Start building today! 🚀 - Blog: t.co/AuiWxugjE6 - API: t.co/aS1QEkS8nu - Token Plan: t.co/YRvcGdB9Bv More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀
Context
China's most downloaded model family now has a closed flagship that independent leaderboards place at the top of Chinese models at a lower cost per task, which adds price pressure on US labs selling similar agent work.
Alibaba's Qwen team announced Qwen4-Max on X at 9:01 PM ET on November 15, which was the morning of November 16 in Hangzhou. TechNode reported that the model had been in a limited preview since November 11 before it became generally available on QwenCloud. Its predecessor, Qwen3.8-Max, used the same 2.4 trillion parameter scale.
Deep dive
Alibaba says Qwen4-Max is the most capable model in the Qwen family so far. According to its press release, it is a sparse mixture of experts model with 2.4 trillion total parameters that activates 71 billion per token, 25 percent fewer active parameters than Qwen3.8-Max at the same scale. It accepts images natively and supports a context window of one million tokens. On X, the Qwen team said it outperforms Qwen3.8-Max with up to about 50 percent fewer tokens and cuts the average cost per task by more than 60 percent in internal testing. API prices are $1.50 per million input tokens, $4.50 per million output tokens and $0.075 for implicit cache hits.
Outside scores arrived within hours. Artificial Analysis said Qwen4-Max scores 43 on its Intelligence Index, 9 points above Qwen3.8-Max and 3 points behind Moonshot's Kimi K3.1 Pro, at about 11 percent lower cost per task ($3.27 against $3.68). Arena said the model debuts at number 14 on its Agent Arena and number 6 in Code Arena WebDev, making Alibaba the top Chinese lab and fifth overall, behind Anthropic, Google, OpenAI and Meta. Bloomberg reported that Alibaba claimed coding results above Anthropic's Claude Fable 5.1.
The model is closed weight. Alibaba said smaller Qwen4 models will be released as open weights. InfoWorld reported that Alibaba filtered offensive hacking data out of training while keeping defensive security data, and quoted analysts who urged buyers to treat the cost claims as vendor benchmarks until independent testing confirms them.
What to expect next
The next milestone is the open weight release of smaller Qwen4 models, which Alibaba has promised but not dated. Developers will compare Qwen4-Max with Kimi K3.1 Pro and Tencent's Hy4 on real agent work, where Artificial Analysis found it still trails Kimi on Terminal-Bench 4.0. Watch also whether US labs answer the price cut, and whether the new US rule on remote access to chips affects how Chinese labs train their next models.
