Tencent WorkBuddy Bench – Agentic Coding Leaderboard (workbuddybench.com)

Tencent has unveiled its WorkBuddy Bench — Agentic Coding Leaderboard, which showcases the competitive landscape of various AI models in coding tasks. Notably, Claude Opus 4.8 emerged as a dominant force, leading five out of eight evaluation categories, including code and web tasks. However, no single model claimed total supremacy, as GLM-5.2 and GPT-5.5 excelled in specific areas, emphasizing a diverse talent pool within the AI landscape. This development is significant for the AI/ML community as it highlights the nuances of model performance across different harnesses, revealing that a model's efficiency can vary significantly by the task framework employed. For instance, while GPT-5.5 achieved top-tier scores with minimal output token costs, including the lowest overall, GLM-5.2's high scores in security came at a steep token price. The findings suggest that not only is model architecture crucial, but the integration and parameters that define these tasks can greatly influence outcomes, opening avenues for further optimization and research in AI-based coding solutions.