用GPT开发了个模型PK擂台，在线一键对比！ 每次模型发布，对比评测总是很麻烦。 用 GPT 5.6 sol 开发一个模型评测对比工具。 不过受限于网站形态，适…

With each new model release, developers and users often face the tedious process of comparison evaluation. Traditional methods require manually writing test cases, running evaluations, and organizing results, which is time-consuming and lacks standardization. On X, Xiangyang Qiaomu (@vista8) stated that to address this pain point, he used GPT 5.6 sol to develop the 'Model PK Arena,' aiming to provide an online one-click comparison tool.

The core function of the Model PK Arena is to allow users to quickly compare outputs of different models on the same task. According to the developer, the tool is built using GPT 5.6 sol and currently supports comparison of simple tasks such as text generation and front-end webpage styling. Users can input a task or select a preset question on the online platform to get outputs from multiple models displayed side by side with one click.

The tool's URL is benchmark.qiaomu.ai. The developer has shared the link on X and publicly released the GitHub repository, encouraging community contributions.

The developer clearly points out current limitations: due to the website format, the tool is only suitable for comparing text outputs and simple front-end tasks. Complex tasks such as multi-turn conversations, code generation, and image understanding are not currently supported. Additionally, all test questions are currently generated by AI, which may not fully cover real-world scenarios.

Future plans include collecting test cases from X netizens' private collections to enrich the question bank and improve the tool's representativeness. This reflects a willingness for community co-construction, but the quality and diversity of test cases remain to be verified.

Although limited in functionality, the emergence of the Model PK Arena lowers the barrier for model comparison, especially for quickly verifying model performance on simple tasks. It encourages users to directly participate in evaluation and promotes transparent model comparison. However, the tool's fairness and accuracy depend on the underlying GPT model and user input, and results should be treated with caution.

For the AI community, the appearance of such tools is a sign of a maturing model ecosystem, but more open-source evaluation standards and community oversight are needed to ensure effectiveness.