Alibabas Qwen3-Coder Outshines DeepSeek in Advanced AI Programming Capabilities

Alibaba has introduced what it describes as «the most advanced AI programming tool» — the open-source AI model Qwen3-Coder.

This neural network is designed for programming-related tasks. It possesses the capability to write code and manage complex workflows, showing significant advancements in various fields.

The company markets this product as being «particularly strong» in agent-based AI tasks, where artificial intelligence can operate independently without human intervention.

According to the data presented, Qwen3-Coder has outperformed its Chinese competitors, including models from DeepSeek and K2 by Moonshot AI, on several metrics. For instance, in the American Mathematical Olympiad, it scored 70.3 points compared to DeepSeek-V3-0324’s 46.6.

In the MultiPL-E programming test, the language model achieved a score of 87.9, while DeepSeek managed 82.2.

In certain areas, this neural network is on par with leading American counterparts, such as Claude by Anthropic and GPT-4 by OpenAI.

The Qwen3-235B-A22B-Instruct-2507-FP8 neural network is an upgrade over the previous Qwen3-235B-A22B-FP8 version, operating in a non-reasoning mode—producing results without step-by-step logical deductions. This enhancement allows for a maximum content processing length of up to 256,000 tokens, accommodating significantly larger texts within a single dialogue.

It is worth noting that in May, Alibaba announced a breakthrough in reducing the training costs for AI models focused on search functionalities.