Alibaba Launches Qwen 3.5-Plus: An Open Source AI Model Rivaling GPT-5.2 and Gemini 3 Pro at 1/18th the Cost

Alibaba launched its new flagship AI model, Qwen 3.5-Plus, on February 16, 2026, amid fierce competition between Chinese tech giants on the eve of the Lunar New Year. The model is available as open source under the Apache 2.0 license and promises performance comparable to the best Western models at a fraction of the price.

Key Highlights

397 billion parameters, with only 17 billion active during inference thanks to a hybrid Sparse MoE + Linear Attention architecture
$0.12 per million tokens — approximately 18 times cheaper than Google's Gemini 3 Pro
201 languages supported, with context windows up to 1 million tokens
Open source under Apache 2.0 on Hugging Face and ModelScope
8.6x faster throughput than Qwen3-Max at 32K context

Benchmarks That Challenge Market Leaders

According to Alibaba's published benchmarks, Qwen 3.5-Plus outperforms OpenAI's GPT-5.2, Anthropic's Claude Opus 4.5, and Google's Gemini 3 Pro across several metrics:

MMLU-Pro: 87.8 — evaluating multi-domain understanding and reasoning
GPQA: 88.4 — measuring advanced science knowledge
IFBench: 76.5 — testing instruction-following accuracy, the highest score among all compared models

These remain Alibaba-published benchmarks and will need independent verification.

An Architecture Built for the Agentic Era

Qwen 3.5 stands out with its hybrid architecture combining Gated DeltaNet and Sparse Mixture-of-Experts (MoE), enabling efficient processing of long sequences while minimizing compute costs.

The model is designed for agentic AI — systems capable of executing complex tasks autonomously. It includes visual agent capabilities for interacting with mobile and desktop applications, along with native multimodal capabilities for text and image processing.

Alibaba claims the model can process up to 2 hours of video content thanks to its one-million-token context window.

A Chinese Market in Full Swing

This launch comes amid a wave of AI model releases from Chinese tech companies. ByteDance, Zhipu AI, MiniMax, and Moonshot AI also unveiled new systems during the same week. The Qwen app had already reached 10 million downloads within one week of its beta launch, but faces competition from ByteDance's Doubao, which has over 200 million users.

What This Means for Developers

The open-source release of Qwen 3.5-Plus under Apache 2.0 represents a significant turning point. Developers, startups, and researchers worldwide can now access a frontier-level model without the costs associated with proprietary APIs. At 0.8 yuan ($0.12) per million tokens via Alibaba Cloud Model Studio, the performance-to-price ratio is hard to beat.

Alibaba also announced the upcoming release of Qwen 3.5-Max, an even more powerful version, along with other variants in the coming weeks.

What's Next

The open-source AI race is heating up. With Qwen 3.5-Plus, Alibaba isn't just competing with the most advanced models — it's making them accessible at a fraction of the cost. For businesses and developers in the MENA region and beyond, this opens new possibilities for integrating cutting-edge AI into their products and services.

Source: Caixin Global