Need suggestions on token usage/caching/optimization
So I have used claude code for an year and recently started using deepseek v4 pro using deepseek official api with claude code terminal app and have been quite impressed with the performance.
However I have noticed deepseek leaves a lot of gaps and misses in implementation when used in a large repo.
Reading through the review of opencode I tool subscription of open code go plan to use models like Minimax m3 and mimo 2.5 pro for code reviews.
I have used deepseek v4 pro using Go plan and i have noticed it burns through a lot more tokens when I used deepseek v4 pro (thinking effort max) than official api.
I want to know how are you guys optimising the token usage on go plan?
So for the context -
When I use official api I burn through $3 max per day but today on the go plan i burned my whole $12 in just 2 hours.
Not sure what would be the reason of such token burn. Need suggestions how can I check if caching or working and diagnose the issue.