Alibaba Releases Game-Changing AI Model Ahead of Lunar New Year

In a strategic move timed with Chinese New Year festivities, Alibaba unveiled Qwen3.5-Plus on February 16, 2026 - an open-source large language model that's rewriting the rules of AI performance. This isn't just another incremental update; it represents a fundamental leap in how efficient AI systems can be.

Revolutionary Architecture Delivers Surprising Efficiency

The numbers tell an impressive story: while containing 397 billion total parameters, Qwen3.5-Plus activates just 17 billion during operation. This selective activation approach allows it to outperform models with trillions of parameters while using 60% less memory. In practical terms? It handles long-context scenarios up to 256K tokens with nineteen times the throughput of previous versions.

Cost Performance That Disrupts the Market

Where Qwen3.5-Plus really turns heads is its price-performance ratio. At just ¥0.8 per million tokens via API, it undercuts Gemini 3 Pro by a staggering factor of eighteen. "We've essentially democratized high-end AI capabilities," explains Dr. Lin Wei, lead architect on the project. "What used to require massive corporate budgets is now accessible to individual developers and startups."

The model breaks new ground technologically too - evolving from pure text processing to true multimodal understanding through innovative pretraining on mixed visual-text tokens.

Benchmark Dominance Across the Board

The proof comes in standardized testing where Qwen3.5-Plus consistently outperforms established leaders:

• 87.8 on MMLU-Pro (surpassing GPT-5.2)
• 88.4 on GPQA (topping Claude 4.5)
• Record-breaking IFBench performance

"It's not just about beating competitors," notes AI researcher Emma Chen from Tsinghua University, "but demonstrating how optimized architectures can achieve more with less computational overhead."

Immediate Availability for Developers

The model has already been deployed across Alibaba's various Qwen platforms, giving developers instant access without waitlists or restrictive licensing - a stark contrast to proprietary alternatives.

Key Points:

• Efficiency breakthrough: Does more with fewer activated parameters
• Cost advantage: API pricing at just ¥0.8/million tokens
• Multimodal leap: Native understanding across text and visuals
• Benchmark leader: Outperforms GPT, Claude and Gemini models
• Open access: Available now across Qwen developer platforms