Devin is capable of responding to text commands, building plans, configuring tools, fixing bugs, deploying applications, and training other AI models.
A benchmark evaluation found Devin correctly resolved 13.86% of issues compared to the previous state-of-the-art's 1.96%. The best previous models could only solve 4.80% of issues, even with exact files provided.
[
add
]
[
|
|
...
]