diff --git a/scripts/collect_responses.py b/scripts/collect_responses.py
index 9d22b45..26f7eb6 100644
--- a/scripts/collect_responses.py
+++ b/scripts/collect_responses.py
@@ -61,7 +61,7 @@ def run_frontier_pass(items: list[dict], out: Path, model: str = KIMI_MODEL, max
             # at 1000 tokens, harder prompts exhausted the budget mid-reasoning and
             # returned empty content (finish_reason="length", content=None). 3000
             # leaves enough room for reasoning + a full visible answer.
-            text, usage = chat(client, model, [{"role": "user", "content": it["prompt"]}], max_tokens=3000)
+            text, usage = chat(client, model, [{"role": "user", "content": it["prompt"]}], max_tokens=8000)
             if not text or not text.strip():
                 raise ValueError("empty response content (likely reasoning-token budget exhaustion)")
         except Exception as e: