top of page




Book Now



 The experiment uses the GPT3.5-turbo-16k version of ChatGPT to simulate multi-agent software development, please refer to the original paper for specific parameter settings, let's see the experimental results below.

Table 1: Statistical analysis of software development in ChatDevelopment, including minimum (Min), maximum (Max) and average (Avg) values for each aspect

• Software developed by ChatDevelopment typically generates 39-359 lines of code, with an average of 131.61 lines. The data shows that ChatDevelopment tends to produce software with relatively small code sizes. This is partly due to the design of object-oriented programming, whose reusability enables code reuse through inheritance, reducing redundancy.

• Figure 6: Duration distribution. The bars in the figure, arranged in descending order, show the distribution of software development run times for different tasks.

• On average, it takes 409.84 seconds to develop small software and interfaces using ChatDevelopment, less than 7 minutes. In contrast, traditional custom software development cycles, even using agile software development methods, typically take 2-4 weeks or even months per cycle.


bottom of page