Qwen 4 Max (Leaked Staging)
Repository-scale software engineering outscoring GPT-6 Sol, multilingual enterprise agents across 120 languages, and low-cost 1M-token processing.
SWE-bench Verified
78.6%
MMLU-Pro
91.4%
MATH-500
95.6%
Context Window
1,000,000 tokens
| Specification | Value |
|---|---|
| Developer / Lab | Alibaba Cloud |
| Parameters Total | 1.2 Trillion (Sparse MoE) |
| Parameters Active | 96 Billion Active Parameters |
| Architecture Type | Sparse Mixture-of-Experts (64 Experts, 8 Active) + Dual-Chunk Attention 1M Context + FP4/FP8 Quantization |
| License | Proprietary Commercial API (Alibaba Cloud Model Studio / Bailian) |
| Input Pricing (per 1M) | $1.20 |
| Output Pricing (per 1M) | $4.80 |
Head-to-Head Comparisons with Qwen 4 Max (Leaked Staging)
Qwen 4 Max (Leaked Staging) vs Gemini 4 Argon →Qwen 4 Max (Leaked Staging) vs GPT-6.1 Sol →Qwen 4 Max (Leaked Staging) vs Gemini 4 Pro (Leaked Benchmark) →Qwen 4 Max (Leaked Staging) vs Space Bunny Alpha (Stealth) →Qwen 4 Max (Leaked Staging) vs GPT-6 Sol →Qwen 4 Max (Leaked Staging) vs GPT-6 Luna →