Alibaba QwenLanguage modelPreview

Qwen3.6-Max

Alibaba's proprietary flagship, tuned hard for coding agents and science code.

Compare head-to-head Race it in 3D

Overall Intelligence

#21

of 56
8.6 pts behind #1

races led

weeks at #1

80.1

score

Released: Apr 20, 2026; 6 mo ago
Context: 256K; 64K max output
Price: $1.6 / $6.4; per 1M in/out · $2.80 blended
Speed: 70 tok/s; 3s to first token
Modalities: text · image; outputs text
Weights: Closed; API and apps

The 30-second version

Alibaba's proprietary flagship (in preview), tuned hard for coding agents and scientific code.

Key takeaways

1Strong on science code and data work.
2Preview status: expect changes.
3Competitive pricing at $1.60 / $6.40.
4256K context, smaller than most rivals.

What you need to know

Purpose

Alibaba's top model for enterprise coding agents and scientific computing in the Qwen ecosystem.

What Qwen optimised for

Agentic coding and science workloads, with strong Chinese and multilingual support.

Unique strengths

Scientific and data code.
Bilingual strength.
Integration with Alibaba Cloud.

Insight

How it differs from rivals. Versus open Qwen models: notably stronger, but closed. Versus GPT-6.1 Sol: cheaper, behind on writing and general reasoning.

Where it ranks

Current snapshot · illustrative scores · click any race to watch it

Qwen3.6-MaxBest on each axis

Overall

21Overall+8.6 pts

Coding

18Coding+9.7 pts 20Frontend & UI+60 Elo 20Backend & APIs+8.9 pts 21Distributed Systems+12.1 pts 17Full-Stack Apps+10.0 pts 21Debugging & Repair+11.1 pts 22Agentic Coding+12.4 pts 20Code Review & Security+9.5 pts 6Data Science & SQL+2.0 pts 17Mobile Apps+8.2 pts 23Terminal & DevOps+13.3 pts 19Competitive Programming+269

Hard Reasoning

20Hard Reasoning+10.3 pts 19Frontier Questions+16.5 pts 19Abstract & Novel Reasoning+14.8 pts 22Logic & Puzzles+8.9 pts 24Planning & Strategy+11.3 pts

Math

21Math+10.7 pts 22Competition Math+8.8 pts 21Research Math+14.6 pts 20Applied & Quantitative+8.0 pts

Science

20Science+10.0 pts 20Graduate Science QA+6.5 pts 18Biology & Medicine+8.1 pts 20Physics & Chemistry+8.6 pts 22Research Assistance+9.7 pts

Writing & Language

18Writing & Language+9.5 pts 27Creative Writing+78 Elo 18Technical Writing+8.3 pts 20Editing & Style+7.7 pts 11Multilingual+4.0 pts 19Instruction Following+7.5 pts 20Conversation & EQ+62 Elo

Multimodal

25Multimodal+20.2 pts 20Image Understanding+10.3 pts 21Documents & Charts+10.3 pts 26Video Understanding+23.1 pts 28Speech & Audio+27.8 pts

Tool Use & Agents

22Tool Use & Agents+11.7 pts 22Tool Use+8.5 pts 24Computer Use+18.0 pts 19Web Research+14.1 pts 24Long-Horizon+22.3 hrs

Long Context

24Long Context+12.7 pts 241M Recall+11.5 pts 23Long Reasoning+10.2 pts

Speed & Cost

23Speed & Cost+37.1 pts 38Output Speed+452 tok/s 22Latency+2.49s 20Cost-Efficiency+37.3 pts

Trust & Factuality

20Trust & Factuality+10.8 pts 18Factuality+8.5 pts 21Honesty & Calibration+9.5 pts

Overall Intelligence over time

#1 frontierQwen3.6-MaxSpark OpenHaiku 5.5

Best use cases

Data and science code

Pandas, numerical methods and analysis notebooks.

Chinese-language products

Assistants and agents serving Chinese-speaking users.

Strengths

Science code
Near the top of our data-and-SQL race.

Weaknesses

Context
256K is limiting for very large repositories.

When to choose it, and when not to

Qwen3.6-MaxoverV4 Pro

You need stronger data and science code and do not need open weights.

Pick something else when…

Production systems that cannot tolerate preview changes.

Update history

The distilled version first; open any release for the deeper breakdown.

The Qwen Max family

Qwen3April 2025 Qwen3-MaxSeptember 2025 Qwen3.6-MaxApril 2026