AI agent built on Alibaba's open-source model tops global benchmark
What's the story
Alibaba Group Holding's open-source Qwen artificial intelligence (AI) model has helped the agentic framework DeepSWE beat its competitors in a recent test. The achievement was announced by the developers of the software platform, Agentica and Together AI. They revealed that DeepSWE, which is based on Alibaba Cloud's third-generation Qwen3-32B large language model (LLM), topped the latest SWEBench-Verified test with 59% accuracy.
AI advancement
DeepSWE trained on Alibaba Cloud's Qwen3-32B model
The success of DeepSWE in the SWEBench-Verified test highlights Alibaba's growing influence in the global open-source community. The framework was developed by Agentica and Together AI, with training conducted on the Qwen3-32B model. This model is part of Alibaba Cloud's third-generation family of AI models. The open-source nature of these models allows developers to access and modify their source code, enhancing their capabilities and fixing any issues that may arise.
Tech explained
What are agentic frameworks?
Agentic frameworks are software platforms that provide the necessary structure, tools, and functionalities to build, deploy, and manage AI agents. These agents can autonomously perform tasks on behalf of a user or another system. They create plans of specific tasks and subtasks using available resources to achieve their goals. The success of DeepSWE demonstrates the potential of these advanced technologies in revolutionizing how we interact with AI systems.