Given Agent mode is here and working well now I think it should be reasonably expected that we get the abililty to define which models get used where, or at least setting the defaults.
Example:
Main Agent: Claude Sonnet | Claude Opus | DeepSeek R1 | o1
File Writer: Claude Sonnet | o3
Review: …
Flow/project/state manager: o1
Further it would be great to let the agent have some flexibility when assigning tasks to models, if it knows a task is simpler it might choose a faster, simpler model for example from a defined list of tradeoffs.
Combined with MCPs for though processes & memory this could be a huge improvement for relatively small effort.
Great suggestion! Currently model selection is limited to choosing one model at a time through the dropdown menu for all Agent tasks. We’ve got this documented at Cursor – Models
While the Agent currently only supports claude-3.5-sonnet, gpt-4o and o3-mini, with deepseek models coming soon, having task-specific model selection would definitely be a nice optimization
I’ll pass this feedback along to the team - being able to define model preferences per task type and having the Agent intelligently choose between allowed models based on complexity could be really powerful
I feel like this will get more optimized over time, but the net time savings of using faster/less complex models automagically for a given sub task will make the entire flow just so much faster. Super exiting to see this working.
