Alibaba launches Qwen3.5 AI model claiming performance gains over Western rivals

Alibaba unveiled its Qwen3.5 artificial intelligence model on Monday, positioning the system as designed for complex autonomous tasks whilst claiming significant cost and performance improvements over major American competitors including GPT-5.2, Claude Opus 4.5, and Gemini 3 Pro.

The Hangzhou-based company released the model as Chinese technology firms intensify competition to attract users in a domestic market dominated by ByteDance’s Doubao chatbot and DeepSeek. Alibaba said Qwen3.5 operates 60 per cent more cheaply than its predecessor and delivers eight times better throughput when processing large workloads, whilst introducing what the firm terms “visual agentic capabilities” allowing the system to independently execute actions across mobile and desktop applications.

“Built for the agentic AI era, Qwen3.5 is designed to help developers and enterprises move faster and do more with the same compute, setting a new benchmark for capability per unit of inference cost,” Alibaba said in a statement. The company published benchmark results showing the model outperforming previous iterations and the three Western systems across several tests, though it made no reference to DeepSeek, whose viral rise prompted Alibaba to release Qwen 2.5-Max last year.

The launch follows ByteDance’s Saturday release of Doubao 2.0, an upgrade to its chatbot commanding nearly 200 million users in China. ByteDance similarly framed its announcement around AI agent capabilities, underscoring how Chinese developers are racing to position their models for autonomous task execution.

Qwen3.5’s release could build on recent momentum for Alibaba in China’s competitive AI landscape. A coupon campaign encouraging direct purchases through the Qwen chatbot earlier this month drove a sevenfold increase in active users, despite technical glitches. The e-commerce giant was among the first to respond to DeepSeek’s breakthrough last year, though it now faces renewed pressure as the startup prepares to release its next-generation model in coming days.

According to Alibaba’s technical documentation, Qwen3.5 uses a hybrid architecture combining linear attention through Gated Delta Networks with sparse mixture-of-experts routing. The 397-billion-parameter model activates only 17 billion parameters per forward pass, expanding language support to 201 languages and dialects from the previous 119. The company also released Qwen3.5-Open-Source featuring a 256,000-token context window, whilst keeping its largest commercial Max-series models proprietary.

Alibaba published performance charts comparing Qwen3.5 against Western rivals. In the GPQA Diamond category measuring graduate-level reasoning, the model scored 88.7, placing third. In IFBench, which evaluates instruction-following accuracy, Qwen3.5 recorded 76.5, outperforming all compared systems.

The flurry of releases from Chinese AI developers marks the first anniversary of DeepSeek’s initial global breakthrough, which triggered a selloff in technology shares. Industry observers expect the startup’s forthcoming announcement to command significant attention from investors and competitors alike.