fix: set temperature=0.0 for DeepSeek models and send reasoning_effor…

agent/auxiliary_client.py

@@ -201,6 +201,11 @@ def _is_arcee_trinity_thinking(model: Optional[str]) -> bool:
     bare = (model or "").strip().lower().rsplit("/", 1)[-1]
     return bare == "trinity-large-thinking"
 
+def _is_deepseek_model(model: Optional[str]) -> bool:
+    """True for DeepSeek V4 models (v4-pro, v4-flash)."""
+    bare = (model or "").strip().lower().rsplit("/", 1)[-1]
+    return bare.startswith("deepseek-v4")
+
 
 # Context window enforced by ChatGPT's Codex OAuth backend for gpt-5.5.
 # The raw OpenAI API and OpenRouter expose 1.05M for the same slug, but the
@@ -250,6 +255,10 @@ def _fixed_temperature_for_model(
         return OMIT_TEMPERATURE
     if _is_arcee_trinity_thinking(model):
         return 0.5
+    if _is_deepseek_model(model):
+        # DeepSeek defaults to 1.0 which is far too hot for agent/tool-use tasks.
+        # 0.0 gives deterministic outputs critical for reliable tool selection.
+        return 0.0
     return None
 
 

plugins/model-providers/deepseek/__init__.py

@@ -83,6 +83,7 @@ def build_api_kwargs_extras(
 
 
 deepseek = DeepSeekProfile(
+    fixed_temperature=0.0,
     name="deepseek",
     aliases=("deepseek-chat",),
     env_vars=("DEEPSEEK_API_KEY",),