Deepseek R1 downgraded to the ground

Discussion
1

man the first days of R1 was fantastic , I use deepseeknV3 Claude 3.5 and R1 on daily basis at least 4-5 hours aday

R1 ISSUES

  • Cut off mid response
  • most of the time gives wrong code
  • sometimes apply changes while still thinking

overall I feel its been %50 downgraded anything it got in one shot in early days now needs 3-4 attempts

and another issue with cursor itself is that if you change from R1 to V3 , V3 will think sometimes if you change to R1 it still uses V3 and does not think

hope we get some answers and fixes other than (its the same)

2 Likes
2

We use the full Deepseek model and don’t downgrade it, so this definitely sounds unusual. Since Claude 3.5 is working well for you, that’s probably your best bet for critical work right now

For the model switching issue, try completely closing and reopening Cursor when switching - this should force a clean slate and prevent any model confusion

3

i hope deep seek r1 works great because it would be good for my plan / execution model. were r1 plans then sonnet executes