Alibaba Cloud has officially launched its latest open artificial intelligence model, Qwen-3.5, designed to compete with top-tier models in the market. This announcement came just before the start of the Chinese Lunar New Year, marking a significant step in Alibaba”s AI ambitions.
The new Qwen-3.5 series was made available initially on Alibaba”s consumer AI application, with an official announcement following on the blogging platform X at approximately 10:00 AM GMT. This release coincides with a wave of updates from various Chinese AI companies, as Beijing seeks to bolster its position in the global AI landscape.
Among its notable features, Qwen-3.5 introduces a unified vision-language foundation that has been trained on trillions of multimodal tokens, allowing it to process text, images, and structured data within a single framework. The development utilized an efficient hybrid architecture, incorporating Gated Delta Networks and sparse Mixture-of-Experts routing to enhance performance by increasing throughput and reducing latency, while also minimizing computing expenses.
Moreover, the model has expanded its linguistic capabilities to include support for 201 languages and dialects, facilitating its global application. A variant called Qwen-3.5-Open-Source was also released, boasting 397 billion parameters and a 256,000-token context window, which is particularly useful for extended reasoning tasks and document processing. Despite this trend towards open-source models, Alibaba has opted to keep its largest Max-series models proprietary to its commercial ecosystem.
The official account of the Alibaba Qwen development team shared performance metrics on X, highlighting comparisons of Qwen-3.5 with competing models such as GPT-5.2, Claude Opus 4.5, and Google”s Gemini 3 Pro. In graduate-level reasoning, measured by the GPQA Diamond category, Qwen-3.5 achieved a score of 88.7, securing the third position among its peers.
In addition, Qwen-3.5 excelled in the IFBench test, which assesses instruction-following accuracy, achieving a score of 76.5—outperforming all other models tested. This performance positions Alibaba”s model closely to proprietary rivals like Gemini, GPT systems, and Claude models, especially in multilingual and agent-based applications.
However, despite the promising launch, Alibaba Group”s stock experienced a decline of 2.93% during global trading, with shares priced at $25,160.00 as of February 13. This drop occurred during a week where the stock fell nearly 6%, as reported by Google Finance. The Chinese equity markets are currently closed for the Lunar New Year holiday, with trading expected to resume on February 24.
The past year has seen a significant push from Chinese tech companies towards developing open-weight AI models, contrasting sharply with the closed development strategies typically employed by companies in Silicon Valley. In the last week alone, several major Chinese AI developers have rolled out updates to their systems, with Alibaba”s DAMO Academy introducing RynnBrain, a distinct AI model aimed at enhancing robotics applications.
