Aider is not using specified editor-model
Issue
I am trying to use openrouter/deepseek/deepseek-r1-0528:free for ask/architect modes and openrouter/deepseek/deepseek-chat-v3-0324:free for making code changes.
I have tried both architect and ask modes, and they both use openrouter/deepseek/deepseek-r1-0528:free for making changes instead of switching to the editor model.
Also, saying do 5 from ask mode doesn't actually do anything, not sure if that's expected but the ask/code workflow docs makes it look like you don't need to manually switch to code mode before saying do.
Version and model info
diff> /settings
Aider v0.85.1
Main model: openrouter/deepseek/deepseek-r1-0528:free with diff edit format
Weak model: openrouter/deepseek/deepseek-chat-v3-0324:free
Git repo: .git with 4 files
Repo-map: using 4096 tokens, files refresh
Config File (/home/vscode/.aider.conf.yml):
model: openrouter/deepseek/deepseek-r1-0528:free
editor-model: openrouter/deepseek/deepseek-chat-v3-0324:free
weak-model: openrouter/deepseek/deepseek-chat-v3-0324:free
ask> what are some small improvements we can make to this script
░█ Waiting for openrouter/deepseek/deepseek-r1-0528:free
Would you like me to elaborate on any specific improvement or implement one of these?
Tokens: 2.5k sent, 2.2k received.
test3.py
ask> /code
Aider v0.85.1 Main model: openrouter/deepseek/deepseek-r1-0528:free with whole edit format Weak model: openrouter/deepseek/deepseek-chat-v3-0324:free Git repo: .git with 4 files Repo-map: using 4096 tokens, files refresh Added test3.py to the chat.
test3.py
do 5
░█ Waiting for openrouter/deepseek/deepseek-r1-0528:free
I think this the intended behavior, but I agree that /code should use the editor model instead.
You can think of this ask/code workflow as a more fluid version of architect mode, but working just with one model the whole time.
/code mode needs more intelligent model to plan changes and apply it as code.
while editor can be a much dumber model which is good at writing code, supplying diffs.
+1 would be great to have the code mode coupled to the editor_model_name. I was searching it after I saw this thread
Use them together:
Sonnet can build multi-step plans,
then orchestrate a team of Haikus to complete subtasks in parallel.
If I understand well, we can't do it in 2 steps (ie /ask and /code) now without switching the main models.