diff --git a/agent/agent_runtime_helpers.py b/agent/agent_runtime_helpers.py
index 5ab2c3d2d2e..ac0ba80d6ee 100644
--- a/agent/agent_runtime_helpers.py
+++ b/agent/agent_runtime_helpers.py
@@ -663,6 +663,25 @@ def recover_with_credential_pool(
         elif status_code in {401, 403}:
             effective_reason = FailoverReason.auth
 
+    if effective_reason == FailoverReason.upstream_rate_limit:
+        # An upstream provider (e.g. DeepSeek behind OpenRouter) is
+        # rate-limiting the aggregator's traffic — the user's credential is
+        # healthy. Do NOT rotate or mark exhausted; let the caller's fallback
+        # path switch to a different model entirely.
+        upstream = (error_context or {}).get("upstream_provider") if error_context else None
+        if upstream:
+            _ra().logger.info(
+                "Upstream provider %s rate-limited via aggregator — skipping "
+                "credential rotation, deferring to fallback chain",
+                upstream,
+            )
+        else:
+            _ra().logger.info(
+                "Upstream aggregator 429 (provider unknown) — skipping "
+                "credential rotation, deferring to fallback chain"
+            )
+        return False, has_retried_429
+
     if effective_reason == FailoverReason.billing:
         rotate_status = status_code if status_code is not None else 402
         next_entry = pool.mark_exhausted_and_rotate(status_code=rotate_status, error_context=error_context)
diff --git a/agent/chat_completion_helpers.py b/agent/chat_completion_helpers.py
index 7a529052e78..aada15f51ed 100644
--- a/agent/chat_completion_helpers.py
+++ b/agent/chat_completion_helpers.py
@@ -1124,7 +1124,7 @@ def try_activate_fallback(agent, reason: "FailoverReason | None" = None) -> bool
     auth resolution and client construction — no duplicated provider→key
     mappings.
     """
-    if reason in {FailoverReason.rate_limit, FailoverReason.billing}:
+    if reason in {FailoverReason.rate_limit, FailoverReason.billing, FailoverReason.upstream_rate_limit}:
         # Only start cooldown when leaving the primary provider.  If we're
         # already on a fallback and chain-switching, the primary wasn't the
         # source of the 429 so the cooldown should not be reset/extended.
@@ -1142,7 +1142,7 @@ def try_activate_fallback(agent, reason: "FailoverReason | None" = None) -> bool
         # provider again.  Guards the cross-turn replay storm in #24996.
         if (
             len(agent._fallback_chain) > 0
-            and reason not in {FailoverReason.rate_limit, FailoverReason.billing}
+            and reason not in {FailoverReason.rate_limit, FailoverReason.billing, FailoverReason.upstream_rate_limit}
         ):
             _existing_cooldown = getattr(agent, "_rate_limited_until", 0) or 0
             agent._rate_limited_until = max(
diff --git a/agent/conversation_loop.py b/agent/conversation_loop.py
index a3677d61ccc..4ca09c6895d 100644
--- a/agent/conversation_loop.py
+++ b/agent/conversation_loop.py
@@ -2920,6 +2920,7 @@ def run_conversation(
                 is_rate_limited = classified.reason in {
                     FailoverReason.rate_limit,
                     FailoverReason.billing,
+                    FailoverReason.upstream_rate_limit,
                 }
                 _is_transport_failure = classified.reason in {
                     FailoverReason.timeout,
@@ -2934,13 +2935,30 @@ def run_conversation(
                     # still recover.  See _pool_may_recover_from_rate_limit
                     # for the single-credential-pool and CloudCode-quota
                     # exceptions.  Fixes #11314 and #13636.
-                    pool_may_recover = _ra()._pool_may_recover_from_rate_limit(
-                        agent._credential_pool,
-                        provider=agent.provider,
-                        base_url=getattr(agent, "base_url", None),
+                    #
+                    # Exception: an upstream-aggregator 429 — the credential
+                    # pool can't help when the *upstream* model (DeepSeek,
+                    # etc.) is throttling OpenRouter, so always fall back to a
+                    # different model regardless of pool state.
+                    _is_upstream = classified.reason == FailoverReason.upstream_rate_limit
+                    pool_may_recover = (
+                        False if _is_upstream
+                        else _ra()._pool_may_recover_from_rate_limit(
+                            agent._credential_pool,
+                            provider=agent.provider,
+                            base_url=getattr(agent, "base_url", None),
+                        )
                     )
                     if not pool_may_recover:
-                        if classified.reason == FailoverReason.billing:
+                        if _is_upstream:
+                            _upstream_name = (classified.error_context or {}).get(
+                                "upstream_provider", "aggregator"
+                            )
+                            agent._buffer_status(
+                                f"⚠️ Upstream {_upstream_name} rate-limited — "
+                                "switching to fallback model..."
+                            )
+                        elif classified.reason == FailoverReason.billing:
                             agent._buffer_status(
                                 "⚠️ Billing or credits exhausted — switching to fallback provider..."
                             )
diff --git a/agent/error_classifier.py b/agent/error_classifier.py
index a64683ba41e..8111880a7ec 100644
--- a/agent/error_classifier.py
+++ b/agent/error_classifier.py
@@ -31,6 +31,9 @@ class FailoverReason(enum.Enum):
     # Billing / quota
     billing = "billing"                  # 402 or confirmed credit exhaustion — rotate immediately
     rate_limit = "rate_limit"            # 429 or quota-based throttling — backoff then rotate
+    # Upstream model rate-limited (aggregator 429) — fallback to a different
+    # model, NOT credential rotation. The user's key is healthy.
+    upstream_rate_limit = "upstream_rate_limit"
 
     # Server-side
     overloaded = "overloaded"            # 503/529 — provider overloaded, backoff
@@ -909,6 +912,22 @@ def _classify_by_status(
                 FailoverReason.overloaded,
                 retryable=True,
             )
+        # Distinguish an OpenRouter-aggregator upstream 429 (an upstream model
+        # like DeepSeek rate-limited OpenRouter's aggregate traffic) from an
+        # account-level 429 (the user's key is actually throttled). OpenRouter
+        # wraps upstream errors with the outer message "Provider returned
+        # error" — the user's key is healthy, so marking it exhausted / rotating
+        # is wrong and burns the key for ~24min. Fall back to a different model.
+        if _is_openrouter_upstream_error(body, provider):
+            upstream_provider = _extract_upstream_provider_name(body)
+            ctx = {"upstream_provider": upstream_provider} if upstream_provider else {}
+            return result_fn(
+                FailoverReason.upstream_rate_limit,
+                retryable=True,
+                should_rotate_credential=False,
+                should_fallback=True,
+                error_context=ctx,
+            )
         return result_fn(
             FailoverReason.rate_limit,
             retryable=True,
@@ -1445,3 +1464,49 @@ def _extract_message(error: Exception, body: dict) -> str:
             return msg.strip()[:500]
     # Fallback to str(error)
     return str(error)[:500]
+
+
+def _is_openrouter_upstream_error(body: Any, provider: str) -> bool:
+    """Detect OpenRouter's aggregator-wrapped upstream provider errors.
+
+    OpenRouter returns errors from upstream model providers (DeepSeek,
+    Anthropic, etc.) wrapped with the outer message "Provider returned error"
+    and the real error nested in ``metadata.raw``. This signal means the
+    user's OpenRouter key is healthy — the upstream provider is the one that
+    failed — so credential rotation is the wrong recovery.
+    """
+    if not isinstance(body, dict):
+        return False
+    provider_lower = (provider or "").strip().lower()
+    err = body.get("error")
+    if not isinstance(err, dict):
+        return False
+    outer_msg = str(err.get("message") or "").strip().lower()
+    if outer_msg != "provider returned error":
+        return False
+    # Require either the explicit OpenRouter provider OR the metadata shape
+    # that only OpenRouter produces (metadata.raw / metadata.provider_name).
+    if provider_lower == "openrouter":
+        return True
+    metadata = err.get("metadata")
+    if isinstance(metadata, dict) and (
+        "raw" in metadata or "provider_name" in metadata
+    ):
+        return True
+    return False
+
+
+def _extract_upstream_provider_name(body: Any) -> Optional[str]:
+    """Pull the upstream provider name out of OpenRouter's error metadata."""
+    if not isinstance(body, dict):
+        return None
+    err = body.get("error")
+    if not isinstance(err, dict):
+        return None
+    metadata = err.get("metadata")
+    if not isinstance(metadata, dict):
+        return None
+    name = metadata.get("provider_name")
+    if isinstance(name, str) and name.strip():
+        return name.strip()
+    return None
diff --git a/scripts/release.py b/scripts/release.py
index a911f20a7ce..1651d8d3a1e 100755
--- a/scripts/release.py
+++ b/scripts/release.py
@@ -50,6 +50,7 @@ AUTHOR_MAP = {
     "phanvanhoa@gmail.com": "theAgenticBuilder",  # PR #14180 salvage (route delegate_task progress lines through _safe_print so ACP stdio JSON-RPC frames stay clean)
     "huangxudong663@gmail.com": "huangxudong663-sys",  # PR #15157 salvage (isinstance(dict) guard on tool-call model_extra; NVIDIA NIM non-dict crash)
     "39369769+jasonQin6@users.noreply.github.com": "jasonQin6",  # PR #15093 salvage (session staleness guard on stream consumer run() loop; #11016 follow-up)
+    "znding04@gmail.com": "znding04",  # PR #15487 salvage (distinguish OpenRouter upstream 429 from account 429; upstream_rate_limit failover reason)
     "cypher@augmentl.com": "Nickperillo",  # PR #8008 salvage (Discord channel-name matching + flush pending sends on shutdown)
     "tenoryang@outlook.com": "MarioYounger",  # PR #9028 salvage (bash/sh heredoc approval, NFKC homograph fold, execute_code CREDS/BEARER/APIKEY env filter)
     "peet.wannasarnmetha@gmail.com": "peetwan",  # PR #51841 salvage (loopback ws-ping tuning + token-frame coalescing + loop heartbeat; #48445/#50005)
diff --git a/tests/agent/test_error_classifier.py b/tests/agent/test_error_classifier.py
index 270c0d8b8d1..16b881861e6 100644
--- a/tests/agent/test_error_classifier.py
+++ b/tests/agent/test_error_classifier.py
@@ -53,6 +53,7 @@ class TestFailoverReason:
     def test_enum_members_exist(self):
         expected = {
             "auth", "auth_permanent", "billing", "rate_limit",
+            "upstream_rate_limit",
             "overloaded", "server_error", "timeout",
             "context_overflow", "payload_too_large", "image_too_large",
             "model_not_found", "format_error",
@@ -1716,4 +1717,110 @@ class TestMultimodalToolContentUnsupported:
         """Make sure the patterns don't false-positive on normal 400s."""
         e = MockAPIError("bad request: missing field 'model'", status_code=400)
         result = classify_api_error(e, provider="openrouter", model="anthropic/claude-sonnet-4")
-        assert result.reason != FailoverReason.multimodal_tool_content_unsupported
+
+
+class TestOpenRouterUpstreamRateLimit:
+    """Distinguish upstream-provider 429 from account-level 429 on OpenRouter.
+
+    When an upstream model (DeepSeek, Anthropic, etc.) rate-limits OpenRouter's
+    aggregate traffic, OpenRouter returns 429 with the outer message "Provider
+    returned error".  The user's key is healthy — we must fall back to a
+    different model, NOT mark the credential exhausted.
+    """
+
+    def test_openrouter_upstream_429_classified_as_upstream_rate_limit(self):
+        """OpenRouter 429 with 'Provider returned error' → upstream_rate_limit."""
+        e = MockAPIError(
+            "Provider returned error",
+            status_code=429,
+            body={
+                "error": {
+                    "message": "Provider returned error",
+                    "code": 429,
+                    "metadata": {
+                        "provider_name": "DeepSeek",
+                        "raw": '{"error":{"message":"Rate limit exceeded"}}',
+                    },
+                }
+            },
+        )
+        result = classify_api_error(e, provider="openrouter", model="deepseek/deepseek-v4-flash")
+        assert result.reason == FailoverReason.upstream_rate_limit
+        assert result.should_rotate_credential is False
+        assert result.should_fallback is True
+        assert result.error_context.get("upstream_provider") == "DeepSeek"
+
+    def test_upstream_429_metadata_shape_without_explicit_provider(self):
+        """metadata.raw shape alone (provider != openrouter literal) still detected."""
+        e = MockAPIError(
+            "Provider returned error",
+            status_code=429,
+            body={
+                "error": {
+                    "message": "Provider returned error",
+                    "metadata": {"raw": '{"error":{"code":429}}'},
+                }
+            },
+        )
+        # provider passed as the slug-form some callers use
+        result = classify_api_error(e, provider="openrouter", model="x")
+        assert result.reason == FailoverReason.upstream_rate_limit
+
+    def test_account_level_429_still_rotates_credential(self):
+        """A real account-level 429 (no upstream wrapper) → rate_limit, rotates."""
+        e = MockAPIError(
+            "Rate limit exceeded: 200 requests per minute",
+            status_code=429,
+            body={
+                "error": {
+                    "message": "Rate limit exceeded: 200 requests per minute",
+                    "code": 429,
+                }
+            },
+        )
+        result = classify_api_error(e, provider="openrouter", model="deepseek/deepseek-v4-flash")
+        assert result.reason == FailoverReason.rate_limit
+        assert result.should_rotate_credential is True
+
+    def test_upstream_wrapper_without_metadata_on_non_openrouter_not_matched(self):
+        """'Provider returned error' alone on a non-openrouter provider → plain rate_limit."""
+        e = MockAPIError(
+            "Provider returned error",
+            status_code=429,
+            body={"error": {"message": "Provider returned error", "code": 429}},
+        )
+        result = classify_api_error(e, provider="anthropic", model="claude-sonnet-4")
+        assert result.reason == FailoverReason.rate_limit
+
+    def test_upstream_provider_name_missing_yields_empty_context(self):
+        """No provider_name in metadata → upstream_rate_limit with empty context."""
+        e = MockAPIError(
+            "Provider returned error",
+            status_code=429,
+            body={
+                "error": {
+                    "message": "Provider returned error",
+                    "metadata": {"raw": '{"error":{"code":429}}'},
+                }
+            },
+        )
+        result = classify_api_error(e, provider="openrouter", model="x")
+        assert result.reason == FailoverReason.upstream_rate_limit
+        assert result.error_context.get("upstream_provider") is None
+
+    def test_overload_429_takes_precedence_over_upstream(self):
+        """A 429 carrying overload language stays overloaded (retry same key)."""
+        e = MockAPIError(
+            "Provider returned error",
+            status_code=429,
+            body={
+                "error": {
+                    "message": "service is temporarily overloaded",
+                    "metadata": {"provider_name": "DeepSeek"},
+                }
+            },
+        )
+        result = classify_api_error(e, provider="openrouter", model="x")
+        # Overload disambiguation runs first; the outer message is the overload
+        # phrase, so this is an overload, not an upstream rate-limit.
+        assert result.reason == FailoverReason.overloaded