A new open-source project, referred to as OpenDevin, developed by John and his team, including Princeton's lab, is making headlines for its performance in software engineering tasks. The project utilizes GPT4 and has shown impressive results on the SWE Bench test set, achieving a 12.29% success rate on 100% of the test set compared to Devin's 13.84% on 25% of the test. This achievement is notable as it suggests the open-source agent is nearly on par with Devin, taking approximately 93 seconds on average to complete tasks. OpenDevin allows users to interact with the source code, providing a more transparent approach to understanding its functionality. The project's ability to handle tasks by viewing only 100 lines at a time has been highlighted as a significant advantage over viewing larger portions of code. The community has praised John and the team for their groundbreaking work.
Thanks to Devin for the contribution to OpenDevin! It's great to see that even AI programmers believe in the power of open source 😃 https://t.co/DrpBKnzdUb
Very nice project OpenDevin an open-source project aiming to replicate Devin, an autonomous AI software engineer who is capable of executing complex engineering tasks and collaborating actively with users on software development projects. However, for local LLM may not be…
The big difference between SWE-Agent and Devin is you can actually try out SWE-Agent, read the source code and understand what it's doing to achieve these results. Hat's off to @jyangballin and the team doing research into this. https://t.co/B2AHrSOdfb
Open source with results close to Devin #ai #coding https://t.co/eNxX1dk8FO
Exciting open source Devin from Princeton https://t.co/2eiigNO0Pn
Huge: Open-source agent from Princeton’s lab claims nearly on-par with Devin while taking ~93s on average! https://t.co/0bXOs87zvL
Open source Devin, with very impressive numbers. From the thread: 'letting SWE-agent only view 100 lines at a time was better than letting it view 200 or 300 lines and much better than letting it view the entire file'. https://t.co/mQsplpewWw
An open source Devin getting 12.29% on 100% of the SWE Bench test set vs Devin's 13.84% on 25% of the test set! This uses GPT4, so imagine what GPT5 could do! Although this works on Github repo issues, I'm assuming the next step is a full generalization! Great work John and team! https://t.co/BQW4mp7koN