(Yicai) July 23 -- Chinese internet giant Alibaba Group Holding has launched the Qwen3-Coder, an open-source artificial intelligence coding model that matches and surpasses some of the most powerful products worldwide.

The Qwen3-Coder's performance on the SWE-Bench, which evaluates models' ability to complete software engineering tasks, is comparable to the latest Claude 4 series from US AI startup Anthropic and outperforms OpenAI's GPT-4.1, Hangzhou-based Alibaba announced today.

Alibaba's new AI model has achieved significant breakthroughs in coding capabilities and agent invocation abilities, the company noted, adding that a novice programmer can use the Qwen3-Coder to complete a week's work of a senior programmer in just a day.

Despite the vast application prospects of AI coding models, ensuring the accuracy of code written for complex projects by such tools remains a challenge, industry insiders told Yicai.

The Qwen3-Coder will soon be integrated into Tongyi Lingma, the AI coding assistant that Alibaba's cloud computing arm launched in October 2023, the parent firm said, adding that the tool is being widely used within the company and by numerous enterprise clients of Alibaba Cloud.

Chinese internet giants are making strides in AI software development, where Cursor, a code editor developed by California-based Anysphere, keeps gaining popularity. Baidu launched its intelligent code assistant Baidu Comate in October 2023, while TikTok's parent company ByteDance unveiled Trae in January.

The penetration rate of AI coding models in the software developers field is around 20 percent, primarily concentrated in frontend and small projects, with their application in complex backend systems still having limitations, a representative from Tongyi Lingma told Yicai in May.

The rate will likely increase to 40 percent to 50 percent within a year, the source added.

Editors: Dou Shicong, Martin Kadiev