Qwen3.5 2B burns all the output tokens while thinking
By adithyaharish · 2026-06-30 · 2 points · 0 comments
I am experimenting with the model and then model spends all its output tokens while thinking making no room left for final output. I have even set thinking budget, but still does not work, anybody has any workarounds or something I am missing?
Open the full discussion on BetterNews