diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
index 63fc58685..5ea5b14df 100644
--- a/gpttype_adapter.cpp
+++ b/gpttype_adapter.cpp
@@ -6657,7 +6657,7 @@ size_t gpttype_save_state_kv(int slot)
             savestates[slot].current_savestate_size   = newsize;
             savestates[slot].savestate_context_tokens = current_context_tokens;
             savestates[slot].media_signature = media_composite_image_signature;
-            float * lgptr = llama_get_logits(llama_ctx_v4);
+            float * lgptr = (draft_is_mtp ? llama_get_logits_ith(llama_ctx_v4, -1) : llama_get_logits(llama_ctx_v4));
             savestates[slot].latest_logits.assign(lgptr,lgptr+n_vocab);
             int maxedpos = llama_memory_seq_pos_max(llama_get_memory(llama_ctx_v4),0);
             //kcpp: so maxedpos appears to always be equal to ctx tokens - 2, if savestate_ctx_tokens > maxedpos + 2 then trim excess