AI News

Qwen4-Max Released: New MoE Architecture, Native Vision, and 1M-Token Context

Ma Dong-seok NLPUpdated Nov 16 21:04

English text is being prepared.

AI summary

Qwen4-Max adopts a new MoE architecture and supports native vision and a 1M-token context. Comparing it with Opus 5.5, the author notes that Qwen4 leads on some Agent capabilities such as DeepSWE, while Opus 5.5 still holds an advantage in complex programming on ProgramBench and complex reasoning on HLE. Compared with Qwen3.8-Max, the new model can reduce token consumption by up to about 50% in internal evaluations, with average task cost dropping by more than 60%.

Why read this

AI score 87

The original article lays out how it divides the work with Opus 5.5 on Agent and complex reasoning tasks, as well as the cost reduction compared with the previous generation.

Topics

Tags

Quoted post

Qwen

@Alibaba_Qwen

📢 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: https://t.co/AuiWxugjE6 - API: https://t.co/aS1QEkS8nu - Token Plan: https://t.co/YRvcGdB9Bv More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀

View original

📢 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: https://t.co/AuiWxugjE6 - API: https://t.co/aS1QEkS8nu - Token Plan: https://t.co/YRvcGdB9Bv More to come! The open weights of smaller Qwen4 models will be released. Stay tuned. 👀

View original post

Body

Qwen4-Max 采用新的 MoE 架构,支持原生视觉和 100 万 Token 上下文。

对比与 Opus 5.5, Qwen4 已经领先部分 Agent 能力,如 DeepSWE。

复杂编程如 ProgramBench 以及 复杂推理HLE ,Opus 5.5 依然保持领先优势。

相较 Qwen3.8-Max,新模型在内部评测中最多可减少约 50% 的 Token 消耗,平均任务成本下降超过 60%。

Source

Ma Dong-seok NLP