Alibaba's open model Qwen3.6 leads Google's Gemma 4 across agentic coding benchmarks

2026-04-17 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, quick

Summary

Alibaba has introduced Qwen3.6-35B-A3B, a new open mixture-of-experts AI model that activates only three of its 35 billion parameters at a time, which Alibaba states reduces compute costs without compromising quality. This model significantly outperforms its predecessor, Qwen3.5-35B-A3B, on agentic coding tasks. Against Google's open Gemma 4-31B, Qwen3.6-35B-A3B leads across all listed coding benchmarks, scoring 73.4 to 52.0 on SWE-bench Verified and 51.5 to 42.9 on Terminal-Bench 2.0. It also shows superior performance on reasoning tests like GPQA (86.0 to 84.3) and AIME26 (92.7 to 89.2), and reportedly matches Claude Sonnet 4.5 on image and video tasks. The model is available through Qwen Studio, Alibaba Cloud Model Studio API (as Qwen3.6 Flash), and for download on Hugging Face and ModelScope.

Key takeaway

For AI Engineers evaluating open models for agentic coding or reasoning tasks, Qwen3.6-35B-A3B presents a compelling option. Its strong benchmark performance against Gemma 4-31B and Qwen3.5, coupled with its cost-efficient mixture-of-experts architecture, suggests it could be a more resource-friendly choice for deployment. You should consider testing its "thinking" and "non-thinking" modes for specific application needs.

Key insights

Alibaba's Qwen3.6-35B-A3B model offers strong performance with reduced compute via a mixture-of-experts architecture.

Principles

• Mixture-of-experts reduces compute costs
• Parameter activation impacts efficiency

In practice

• Try Qwen3.6 in Qwen Studio
• Access via Alibaba Cloud Model Studio API
• Download weights from Hugging Face

Topics

• Alibaba Qwen3.6-35B-A3B
• Mixture-of-Experts Model
• Agentic Coding Benchmarks
• Google Gemma 4
• AI Model Performance

Best for: AI Engineer, AI Architect, Research Scientist, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Qwen3.6-27B beats much larger predecessor on most coding benchmarks — Alibaba's Qwen3.6-27B model offers superior coding performance with significantly fewer parameters than its predecessor.
• Google’s Gemma 4: Is it the Best Open-Source Model of 2026? — Google's Gemma 4 family offers versatile, open-weight LLMs optimized for diverse environments, from edge devices to consumer GPUs.
• Alibaba's open Qwen 3.5 takes aim at GPT-5 mini and Claude Sonnet 4.5 at a fraction of the cost — Alibaba's Qwen 3.5 models achieve competitive performance against top Western LLMs at lower computational cost.
• Gemma 4 31B vs Qwen3.5 27B: Inference Speed, Token-Efficiency, Accuracy, and Memory Consumption — Gemma 4 31B generally outperforms Qwen3.5 27B in accuracy and consistency, with better generalization.
• Qwen3.6 Local Test | Can it Beat Gemma 4? | Coding, OCR, Image Understanding with llama.cpp | 🔴 Live — Qwen 3.6 offers strong agentic coding and visual understanding, but its verbose reasoning can impact efficiency.
• Alibaba's Qwen 3.5 397B-A17 beats its larger trillion-parameter model — at a fraction of the cost — Sparse Mixture-of-Experts models can surpass larger predecessors in performance and efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.