Nov 21, 08:47 PM

Alibaba Launches Qwen2.5 Coding Models, Rivals GPT-4 with Aider 73.7 and McEval 65.9 Scores

The Qwen2.5 series of coding models from Alibaba has been launched, showcasing significant advancements in code generation and repair capabilities. The Qwen2.5-Coder-7B model is touted to rival OpenAI's GPT-4 in coding tasks, achieving high scores on various benchmarks, including 73.7 on Aider and 65.9 on McEval. Additionally, the Qwen2.5-32B model has been highlighted for its superior performance compared to other leading models, including GPT-4o and Claude. The models are now available on multiple platforms, including Bakery and Baseten, with the Qwen2.5-Coder-32B being recognized as the first open-source coding model to match GPT-4o's capabilities. The integration of these models into frameworks such as OpenLLM and CAMEL further enhances their accessibility for developers.