diff --git a/run_agent.py b/run_agent.py
index 720bc19dc..482878019 100644
--- a/run_agent.py
+++ b/run_agent.py
@@ -5816,6 +5816,17 @@ class AIAgent:
             return primary_client
         with self._openai_client_lock():
             request_kwargs = dict(self._client_kwargs)
+        # Per-request OpenAI-wire clients (used by both the non-streaming
+        # chat-completions path and the streaming chat-completions path
+        # in `_interruptible_api_call`) should not run the SDK's built-in
+        # retry loop: the agent's outer loop owns retries with credential
+        # rotation, provider fallback, and backoff that the SDK can't
+        # see. Leaving SDK retries on (default 2) compounds with our outer
+        # retries and lets a single hung provider request stretch to ~3x
+        # the per-call timeout before our stale detector reports it.
+        # Shared/primary clients and Anthropic / Bedrock paths are
+        # unaffected (they don't go through here).
+        request_kwargs["max_retries"] = 0
         if (
             base_url_host_matches(str(request_kwargs.get("base_url", "")), "api.githubcopilot.com")
             and self._api_kwargs_have_image_parts(api_kwargs or {})