diff --git a/litellm/integrations/websearch_interception/handler.py b/litellm/integrations/websearch_interception/handler.py
index 587da997f94..ba701ba1ef2 100644
--- a/litellm/integrations/websearch_interception/handler.py
+++ b/litellm/integrations/websearch_interception/handler.py
@@ -1794,11 +1794,9 @@ class WebSearchInterceptionLogger(CustomLogger):
             k: v for k, v in kwargs.items() if not k.startswith("_websearch_interception") and k not in internal_params
         }
 
-        full_model_name = model
-        if "custom_llm_provider" in kwargs:
-            custom_llm_provider: Final = kwargs["custom_llm_provider"]
-            if not model.startswith(custom_llm_provider) and "/" not in model:
-                full_model_name = f"{custom_llm_provider}/{model}"
+        from litellm.litellm_core_utils.core_helpers import qualify_provider_stripped_model
+
+        full_model_name: Final = qualify_provider_stripped_model(model, kwargs.get("custom_llm_provider", ""))
 
         verbose_logger.debug(
             "WebSearchInterception: Built chat completion request patch model=%s messages=%d",
diff --git a/litellm/litellm_core_utils/core_helpers.py b/litellm/litellm_core_utils/core_helpers.py
index aa7d6ca1699..cc7fab718f3 100644
--- a/litellm/litellm_core_utils/core_helpers.py
+++ b/litellm/litellm_core_utils/core_helpers.py
@@ -29,6 +29,20 @@ def is_codex_user_agent(user_agent: str) -> bool:
     return bool(_CODEX_CLIENT_PREFIX_RE.match(user_agent))
 
 
+def qualify_provider_stripped_model(model: str, custom_llm_provider: str) -> str:
+    """
+    Put the provider prefix back on a model an agentic follow-up re-dispatches with.
+
+    Handlers are handed the provider-stripped model, and for providers that route through
+    a sub-path (``bedrock/mantle/...``, ``openrouter/openai/...``) that remainder still
+    holds a slash. Treating any slash as "already qualified" drops the prefix and leaves a
+    string no provider can be resolved from.
+    """
+    if not custom_llm_provider or model.startswith(f"{custom_llm_provider}/"):
+        return model
+    return f"{custom_llm_provider}/{model}"
+
+
 def safe_divide_seconds(seconds: float, denominator: float, default: float | None = None) -> float | None:
     """
     Safely divide seconds by denominator, handling zero division.
diff --git a/litellm/llms/custom_httpx/llm_http_handler.py b/litellm/llms/custom_httpx/llm_http_handler.py
index 7109e6942d1..593edf77c5b 100644
--- a/litellm/llms/custom_httpx/llm_http_handler.py
+++ b/litellm/llms/custom_httpx/llm_http_handler.py
@@ -5599,9 +5599,9 @@ class BaseLLMHTTPHandler:
         if patch.messages is None:
             raise ValueError("Agentic loop plan missing patched messages")
 
-        full_model_name = patch.model or model
-        if "/" not in full_model_name:
-            full_model_name = f"{custom_llm_provider}/{full_model_name}"
+        from litellm.litellm_core_utils.core_helpers import qualify_provider_stripped_model
+
+        full_model_name: Final = qualify_provider_stripped_model(patch.model or model, custom_llm_provider)
 
         optional_params_for_followup: Final = dict(optional_params)
         optional_params_for_followup.update(patch.optional_params)
diff --git a/tests/test_litellm/litellm_core_utils/test_core_helpers.py b/tests/test_litellm/litellm_core_utils/test_core_helpers.py
index e937be47441..b1575cadc26 100644
--- a/tests/test_litellm/litellm_core_utils/test_core_helpers.py
+++ b/tests/test_litellm/litellm_core_utils/test_core_helpers.py
@@ -11,6 +11,7 @@ from litellm.litellm_core_utils.core_helpers import (
     get_or_create_metadata_bucket,
     map_finish_reason,
     normalize_drop_params,
+    qualify_provider_stripped_model,
     reconstruct_model_name,
     redact_nested_match_and_regex_keys,
 )
@@ -406,3 +407,43 @@ class TestIsExpectedClientError:
             category=RateLimitErrorCategory.VENDOR_RATE_LIMIT,
         )
         assert is_expected_client_error(vendor_limit) is False
+
+
+class TestQualifyProviderStrippedModel:
+    """#38829: an agentic follow-up re-dispatched the provider-stripped model. For providers
+    that route through a sub-path the remainder still holds a slash, and the old
+    "any slash means already qualified" test dropped the prefix, leaving a string
+    litellm.acompletion could not resolve a provider from."""
+
+    @pytest.mark.parametrize(
+        "model,provider,expected",
+        [
+            # the reported case: bedrock's OpenAI-compatible sub-path
+            ("mantle/anthropic.claude-sonnet-5", "bedrock", "bedrock/mantle/anthropic.claude-sonnet-5"),
+            ("invoke/anthropic.claude-v2", "bedrock", "bedrock/invoke/anthropic.claude-v2"),
+            # another provider whose stripped model keeps a slash
+            ("openai/gpt-4o", "openrouter", "openrouter/openai/gpt-4o"),
+            # the ordinary case still works
+            ("gpt-4o", "openai", "openai/gpt-4o"),
+            ("claude-sonnet-4-5", "anthropic", "anthropic/claude-sonnet-4-5"),
+        ],
+    )
+    def test_the_provider_prefix_is_restored(self, model, provider, expected):
+        assert qualify_provider_stripped_model(model, provider) == expected
+
+    @pytest.mark.parametrize(
+        "model,provider",
+        [
+            ("bedrock/mantle/anthropic.claude-sonnet-5", "bedrock"),
+            ("openai/gpt-4o", "openai"),
+        ],
+    )
+    def test_an_already_qualified_model_is_left_alone(self, model, provider):
+        assert qualify_provider_stripped_model(model, provider) == model
+
+    def test_a_provider_that_only_shares_a_prefix_is_still_qualified(self):
+        """`openai` must not be read as a prefix of `openai_like`."""
+        assert qualify_provider_stripped_model("openai_like/foo", "openai") == "openai/openai_like/foo"
+
+    def test_no_provider_leaves_the_model_untouched(self):
+        assert qualify_provider_stripped_model("gpt-4o", "") == "gpt-4o"