Alibaba takes 2.93% hit despite bullish benchmarks from Qwen-3.5 AI model release

Alibaba Cloud has launched Qwen-3.5, its next-generation open artificial intelligence model, which the company claims can compete “with state-of-the-art leading models.” 

On the eve of the Chinese Lunar New Year, the Hangzhou-based tech company unveiled the latest iteration of its AI large language model Qwen. The new 3.5-series models were made available first on Alibaba’s flagship consumer AI app on Monday afternoon, followed by an official announcement made on social platform X at around 10:00 AM GMT.

🚀 Qwen3.5-397B-A17B is here: The first open-weight model in the Qwen3.5 series.

🖼️Native multimodal. Trained for real-world agents.
✨Powered by hybrid linear attention + sparse MoE and large-scale RL environment scaling.
⚡8.6x–19.0x decoding throughput vs Qwen3-Max
🌍201… pic.twitter.com/Pq0qIk54MB

— Qwen (@Alibaba_Qwen) February 16, 2026

The launch comes during a week in which several Chinese AI firms released the new versions of their flagship AI systems, as Beijing grows its ambitions to surpass the West in AI capabilities.

Alibaba Group’s shares fell 2.93% in global trading following the announcement, with the stock quoted at $25,160.00 as of February 13. The decline also extended to a five-day drop of nearly 6%, according to Google Finance data. China’s equity markets are closed for the Lunar New Year holiday from February 16 to February 23, with trading set to resume on February 24.

Qwen-3.5 architecture sets sights on multimodal AI expansion

According to Alibaba Cloud’s GitHub notes, Qwen-3.5 includes a unified vision-language foundation trained on trillions of multimodal tokens. The system can now process text, images, and structured inputs within a single framework.

During development of the new LLM version, the company used an efficient hybrid architecture and Gated Delta Networks with sparse Mixture-of-Experts routing to increase throughput, limit latency, and reduce computing costs. The model has also increased its linguistic support to 201 languages and dialects to expand its deployment worldwide.

A variant named Qwen-3.5-Open-Source was also released, featuring 397 billion parameters and a 256,000-token context window for longer reasoning chains and document processing. Despite releasing most of its AI LLM models as open source, Alibaba continues to keep its largest Max-series models closed for its commercial ecosystem.

The official X account of the Alibaba Qwen development team posted performance charts alongside the announcement, comparing Qwen-3.5 to GPT-5.2, Claude Opus 4.5, and Google’s Gemini 3 Pro.

In the GPQA Diamond category, which measures graduate-level reasoning, Qwen-3.5 scored 88.7, coming third among the other LLMs. In IFBench, a test evaluating instruction-following accuracy, Qwen-3.5 recorded a score of 76.5, outperforming all the other models.

The results place Alibaba’s model within striking distance of proprietary competitors such as Gemini, GPT-series systems, and Claude models in multilingual and agent-based workloads.

China taps into open-source AI to compete with the US

Over the past year, Chinese technology firms have doubled down on creating open-weight models, which is much different from the closed development approaches seen in Silicon Valley companies.

Nearly every major Chinese AI developer introduced updated systems in the past seven days, with Alibaba’s DAMO Academy launching RynnBrain, a separate AI model aimed at robotics applications. The company’s RynnBrain demonstrations featured a robot equipped with pincers that counted oranges, placed them in a basket, and retrieved milk from a refrigerator.

“One of its key innovations is built-in time and space awareness,” said Adina Yakefu, a researcher at Hugging Face.

On February 12, TikTok’s parent company, ByteDance, debuted Seedance 2.0, a model capable of generating realistic videos from text prompts, images, or existing clips. On social media, users have been sharing samples of AI-generated videos featuring characters from Marvel comic movies, like Thanos.

As reported by Cryptopolitan, Marvel rights owner Disney sent a cease-and-desist letter on Friday to ByteDance, accusing the firm of training the system on a “pirated library” of copyrighted characters.

On Monday, ByteDance responded that it “respects intellectual property rights and has heard the concerns regarding Seedance 2.0.”

“We are taking steps to strengthen current safeguards as we work to prevent the unauthorised use of intellectual property and likeness by users,” the company told reporters.

If you're reading this, you’re already ahead. Stay there with our newsletter.