Qwen4-Max Launches with 2.4T Parameters, Native Vision, and 1M Token Context
·2026.11.16 09:00
Key point
Qwen4-Max launches with a 2.4T parameter backbone, native vision, and 1M token context, improving coding and long-horizon task performance over Qwen3.8-Max.
Details
- Qwen4-Max is introduced by Qwen, featuring a 2.4T parameter backbone, native vision, and a 1M token context window.
- Work360 benchmark score increased to 52.69 (from 47.74 in Qwen3.8-Max). The average cost per task dropped 62% from $4.43 to $1.67, driven by API price reductions and token efficiency gains.
- Token usage reduced by approximately 24% (Xhigh effort) and 35% (High effort) on Work360, and by 48-49% on Terminal-Bench 4.0.
- Architecture includes 71B active parameters per token, 328B N-gram Embedding parameters, and GDN + Qwen Sparse Attention (QSA) for efficient long-sequence processing.
Performance & Benchmarks
- Self-evaluation shows Qwen4-Max-Base achieves the lowest bits per byte (BPB) across 5 checkpoints, with BPB 5-13% lower than Qwen3.8-Max-Base and 25% fewer active parameters.
- Last Secure Code Bench function-level pass rate improved from 22.0% to 31.8%, and safe code generation ratio nearly doubled from 8% to 14.7%.
Industry Applications
- Automotive manufacturing (SANF, Pacific Union): Reduced manual programming time for single workpieces by 95% (from 475 minutes to 24 minutes) via end-to-end production line planning and PLC virtual commissioning.
- Cloud resource scheduling (Alibaba Cloud): Achieved a 10.42% reduction in data annotation model call costs, a 4.62% increase in online GPU idle capacity duration, and a 14.61 times improvement in solving speed for reaching the optimal objective in offline CPU rescheduling using MindOpt solver collaboration.
- Scientific discovery: Qwen4 autonomously designed and synthesized a new coolant molecule for data centers via closed-loop dry/wet lab integration on Deep Principle's Mira platform.
Availability & Safety
- Available on QwenCloud (Token Plan, API), QwenWork, Qoder, and QwenStudio.
- Safety governance includes enhanced secure-coding capabilities, removal of offensive cybersecurity data, and strict prohibitions on CBRN weapon development and privacy violations.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.