DeepSeek V4.1 Flash 1017 in OpenCode: fewer loops, usable vision
OpenCode testing of 1017 finds fewer loops and less wasted tokens per task, usable vision for screenshots and UI bugs, and a free endpoint that slows under load versus Go.
a few more things i’ve noticed with DeepSeek V4.1 Flash 1017 in OpenCode: 1. the list price didn’t really change, but the model loops less. in practice that means fewer wasted tokens per completed task, so the cost per result feels lower than the previous checkpoint. 2. vision is actually useful now. i’m much more comfortable handing it screenshots, UI states and visual bugs instead of switching models just for that part of the workflow. 3. the free endpoint and Go don’t feel identical. free can get noticeably slower under load, so for longer coding sessions i’d rather stay on Go. 4. i’ve been able to run it for basically a full working day during off-peak windows. for heavy implementation work, the combination of speed + Go limits is probably the biggest reason i keep coming back to it. it wastes less time getting to the same result.
This post quotes another post: “Hands-on: DeepSeek V4.1 Flash 1017 in OpenCode, five observations”