This commit is contained in:
Paul Gauthier 2024-05-13 11:03:12 -07:00
parent 5b3e9d7b05
commit 45ea2d5871

View file

@ -21,6 +21,13 @@ GPT-4o tops the aider LLM code editing leaderboard at 72.9%, versus 68.4% for Op
GPT-4o did much better than the 4-turbo models, and seems *much* less lazy.
GPT-4o is also able to use aider's established "diff" edit format that uses
`SEARCH/REPLACE` blocks.
This diff format is used by all the other capable models, including Opus and
the original GPT-4 models
The GPT-4 Turbo models have all required the "udiff" edit format, due to their
tendancy to lazy coding.
## Code editing leaderboard