Qwen3.7 Max
Parameters
Undisclosed
Context Length
1M
Modality
Text
Architecture
Dense
License
Proprietary
Release Date
20 May 2026
Knowledge Cutoff
-
API Pricing (per 1M)
Input: $1.25 · Output: $3.75
Evaluation Benchmarks
Rank
#79
| Benchmark | Score | Rank |
|---|---|---|
Professional Knowledge | 0.896 | 🥈 2 |
Graduate-Level QA | 0.924 | 12 |
Reasoning | 0.83 | 32 |
General Text | 1473 | 32 |
Data Analysis | 0.72 | 34 |
General | 0.73 | 36 |
LiveBench Average | 0.73 | 36 |
Coding | 0.74 | 38 |
Web Development | 1517 | 38 |
Mathematics | 0.85 | 40 |
Agentic Coding | 0.44 | 41 |
Coding Index | 0.66 | 57 |
Agentic Index | 0.24 | 73 |
Intelligence Index | 0.30 | 76 |
Rankings
Overall Rank
#79
Coding Rank
#56
About Qwen3.7 Max
Alibaba's proprietary frontier model for the agent era, released May 20, 2026. Qwen3.7-Max is purpose-built as a versatile agent foundation capable of writing and debugging code, automating office workflows, and sustaining autonomous execution across hundreds to thousands of steps. Achieved a 10x geometric mean speedup on an Extend Attention kernel on previously unseen hardware over 35 hours of continuous autonomous execution. Features a 1M token context window and cross-scaffold generalization across Claude Code, OpenClaw, Qwen Code, and other frameworks.
Technical Specifications
Architecture specifications are undisclosed for proprietary models.
Attention
Attention Structure
Multi-Head Attention
Attention Heads
Undisclosed
Key-Value Heads
Undisclosed
Attention Head Dimension
Undisclosed
Position Embedding
Absolute Position Embedding
RoPE Theta
Undisclosed
Sliding Window Attention
Undisclosed
Sliding Window Size
Undisclosed
Sliding Window Ratio
Undisclosed
Linear Attention
Undisclosed
Linear Attention Ratio
Undisclosed
Normalization
Undisclosed
Activation Function
Undisclosed
Dimensions
Hidden Dimension Size
Undisclosed
Number of Layers
Undisclosed
FFN Intermediate Size (Dense)
Undisclosed
Multi-Token Prediction Heads
Undisclosed
Tokenizer
Vocabulary Size
Undisclosed
Resources
About Qwen 3.7
Alibaba's Qwen 3.7 generation is designed for the agent era, delivering frontier-level agentic reasoning and long-horizon autonomous execution. Qwen3.7 models combine deep coding agent capabilities with broad cross-scaffold generalization, sustaining productive execution over multi-hour sessions with thousands of tool calls. The family includes both text-focused and full multimodal variants.