Alibaba Enters the Agentic Era: Meet the All-New Qwen 3.5
Alibaba Enters the Agentic Era: Meet the All-New Qwen 3.5
Alibaba Cloud literally just launched Qwen 3.5 today (February 16, 2026), timing the release for the eve of the Lunar New Year.
This update is a major leap forward, focusing on what Alibaba calls the "Agentic AI Era." Here is the breakdown of what is new:
1. Breakthrough Performance & Cost
Qwen 3.5 isn't just a slight upgrade; it’s designed to be much more efficient than the previous Qwen 3 and 2.5 models.
60% Cheaper: Alibaba has slashed API costs, making it roughly 1/18th the price of competitors like Gemini 3 Pro.
8x More Capable: It is significantly better at handling massive, complex workloads compared to Qwen 3.
Benchmarking: Early reports claim Qwen 3.5-Plus outperforms or rivals GPT-5.2, Claude 4.5, and Gemini 3 Pro in reasoning and coding.
2. "Visual Agentic" Capabilities
This is the "killer feature" of 3.5. Unlike older models that just "chat," Qwen 3.5 is built to act.
Multimodal Integration: It was pre-trained on a mix of text and visual data from the start.
App Navigation: It can independently execute tasks across mobile and desktop applications (e.g., "Find the cheapest flight on this app and draft an itinerary").
Native Multi-Step Planning: Instead of generating word-by-word, the model can "plan" several steps ahead, which nearly doubles its response speed for complex tasks.
3. New Architecture: Sparse MoE
Qwen 3.5 uses an extremely sparse Mixture-of-Experts (MoE) architecture.
Total Parameters: 397 Billion.
Active Parameters: It only "wakes up" about 5% of its brain (roughly 17–20 billion parameters) for any given task. This allows it to have the "knowledge" of a massive model but the "speed and cost" of a small one.
Context Window: It supports a 1 million token context window, allowing you to upload entire libraries of code or books for analysis.
Key Models Released Today:
| Model | Focus |
| Qwen 3.5-Plus | The flagship high-performance model for complex reasoning. |
| Qwen 3.5-9B/35B | Open-source versions optimized for local developer use. |
| Qwen 3.5-Coder | A specialized version that achieves 70%+ on the SWE-Bench (coding) tests. |
The race for AI supremacy just shifted from "chatting" to "doing." Today, Alibaba Cloud officially unveiled Qwen 3.5, its most powerful and efficient flagship AI model to date.
Built specifically for the "Agentic Era," Qwen 3.5 isn't just about answering questions; it’s designed to execute complex tasks independently, acting as a bridge between human intent and digital action.
What Makes Qwen 3.5 Different?
While previous models focused on being great conversationalists, Qwen 3.5 is an AI Agent first. It has been optimized to navigate apps, manage workflows, and handle multi-step reasoning with minimal human intervention.
Here are the key highlights from the launch:
Native Multimodal Intelligence: Qwen 3.5 is a native vision-language model. This means it doesn't just "see" an image; it understands the UI of a mobile or desktop app, allowing it to click buttons and navigate software—a feature Alibaba calls "Visual Agentic Capabilities."
The Power of 397B (with a twist): The flagship model, Qwen3.5-397B, uses a sparse Mixture-of-Experts (MoE) architecture. While it has a massive 397 billion parameters, it only activates 17 billion per request. This gives it "God-tier" intelligence at a fraction of the computational cost.
Massive Context Window: By default, Qwen 3.5 supports a 1 million token context window, allowing it to process entire libraries of documents or hours of video in one go.
Global Reach: Language support has expanded from 119 to 201 languages, making it one of the most linguistically diverse models in the world.
Performance vs. The Giants
Alibaba isn't playing it safe. In their announcement, they shared benchmarks showing Qwen 3.5 performing on par with—or exceeding—top-tier models like GPT-5.2, Claude 4.5 Opus, and Gemini 3 Pro.
Most impressively, Alibaba claims Qwen 3.5 is 60% cheaper to operate and 8 times more efficient at processing large workloads than the previous Qwen 2.5 version. For developers, this means more power for less money.
The "AI That Acts"
We are seeing a shift in the Qwen App. Users in China are already testing features where the AI can:
Order food and complete payments via Alipay without leaving the chat.
Plan entire travel itineraries and book the flights/hotels autonomously.
Extract data from 100+ documents simultaneously to build a research report.
Open Source Commitment
True to their "Open-Source" philosophy, Alibaba has already released the open-weight version of Qwen3.5-397B-A17B on
Final Thoughts: Qwen 3.5 marks a turning point where AI stops being a search engine replacement and starts becoming a digital employee. Whether you are a developer looking for an affordable API or a user wanting a personal assistant that actually does the work, Qwen 3.5 is a model you cannot ignore.
Want to try it yourself? You can experience the new capabilities on the
or via Alibaba Cloud’s Model Studio. official Qwen Chat site
No comments