diff --git a/litellm/__init__.py b/litellm/__init__.py
index ede8a73453d..5a461801b62 100644
--- a/litellm/__init__.py
+++ b/litellm/__init__.py
@@ -546,7 +546,7 @@ _key_management_system: Optional["KeyManagementSystem"] = None
 #### PII MASKING ####
 output_parse_pii: bool = False
 #############################################
-from litellm.litellm_core_utils.get_model_cost_map import get_model_cost_map
+from litellm.litellm_core_utils.get_model_cost_map import get_model_cost_map, mark_litellm_import_complete
 
 model_cost = get_model_cost_map(url=model_cost_map_url)
 cost_discount_config: Dict[str, float] = {}  # Provider-specific cost discounts {"vertex_ai": 0.05} = 5% discount
@@ -2405,3 +2405,5 @@ def __getattr__(name: str) -> Any:
 
 
 # ALL_LITELLM_RESPONSE_TYPES is lazy-loaded via __getattr__ to avoid loading utils at import time
+
+mark_litellm_import_complete()
diff --git a/litellm/litellm_core_utils/get_model_cost_map.py b/litellm/litellm_core_utils/get_model_cost_map.py
index cdc4810ff04..f81ddbfee2e 100644
--- a/litellm/litellm_core_utils/get_model_cost_map.py
+++ b/litellm/litellm_core_utils/get_model_cost_map.py
@@ -13,6 +13,7 @@ import hashlib
 import json
 import os
 import random
+import threading
 import time
 from collections.abc import Awaitable, Callable
 from dataclasses import dataclass, replace
@@ -176,6 +177,11 @@ class GetModelCostMap:
 RETRYABLE_FETCH_STATUS_CODES: Final = frozenset({429, 500, 502, 503, 504})
 MODEL_COST_MAP_FETCH_MAX_ATTEMPTS: Final = 3
 MODEL_COST_MAP_FETCH_MAX_WAIT_SECONDS: Final = 30.0
+_litellm_import_complete = threading.Event()
+
+
+def mark_litellm_import_complete() -> None:
+    _litellm_import_complete.set()
 
 
 @dataclass(frozen=True, slots=True)
@@ -314,12 +320,13 @@ async def _fetch_remote_model_cost_map_with_retry(
 def _fetch_remote_model_cost_map_with_retry_sync(
     url: str,
     timeout: int,
-    max_attempts: int,
+    attempts: range,
     sleep: Callable[[float], None],
     rng: random.Random,
     client: _SyncGetClient,
 ) -> ModelCostMapReloadResult:
-    for attempt in range(1, max_attempts + 1):
+    max_attempts: Final = attempts.stop - 1
+    for attempt in attempts:
         outcome = _attempt_fetch_sync(client=client, url=url, timeout=timeout)
         if not isinstance(outcome, _FetchAttemptRetryable):
             return outcome
@@ -520,6 +527,68 @@ def _finalize_loaded_model_cost_map(loaded: ModelCostMapReloaded) -> ModelCostMa
     return replace(loaded, model_cost_map=_finalize_model_cost_map(loaded.model_cost_map))
 
 
+def adopt_model_cost_map(
+    new_model_cost_map: dict,  # mutable-ok: public API preserves the mutable cost-map contract
+) -> int:
+    import litellm
+    from litellm import utils
+
+    litellm.model_cost = new_model_cost_map
+    utils._invalidate_model_cost_lowercase_map()  # pyright: ignore[reportPrivateUsage]  # required cache invalidation
+    litellm.add_known_models(model_cost_map=new_model_cost_map)
+    fetched_model_count: Final = len(new_model_cost_map) if new_model_cost_map else 0
+    utils.reapply_runtime_model_cost_registrations()
+    return fetched_model_count
+
+
+def _retry_remote_fetch_in_background(
+    url: str,
+    timeout: int,
+    max_attempts: int,
+    sleep: Callable[[float], None],
+    rng: random.Random,
+    client: _SyncGetClient,
+    first_outcome: _FetchAttemptRetryable,
+) -> None:
+    try:
+        first_wait: Final = _next_retry_wait(outcome=first_outcome, attempt=1, max_attempts=max_attempts, rng=rng)
+        if isinstance(first_wait, ModelCostMapReloadUnavailable):
+            return
+        sleep(first_wait)
+        result: Final = _fetch_remote_model_cost_map_with_retry_sync(
+            url=url,
+            timeout=timeout,
+            attempts=range(2, max_attempts + 1),
+            sleep=sleep,
+            rng=rng,
+            client=client,
+        )
+        if isinstance(result, ModelCostMapReloadUnavailable):
+            verbose_logger.warning(
+                "LiteLLM: Failed to fetch remote model cost map from %s after %d attempts; keeping local backup",
+                url,
+                max_attempts,
+            )
+            return
+        _litellm_import_complete.wait()
+        if not GetModelCostMap.validate_model_cost_map(
+            fetched_map=result.model_cost_map,
+            backup_model_count=GetModelCostMap._get_backup_model_count(),  # pyright: ignore[reportPrivateUsage]  # integrity cache
+        ):
+            verbose_logger.warning(
+                "LiteLLM: Fetched model cost map failed integrity check. Using local backup instead. url=%s",
+                url,
+            )
+            return
+        finalized: Final = _finalize_loaded_model_cost_map(result).model_cost_map
+        _cost_map_source_info.source = "remote"
+        _cost_map_source_info.fallback_reason = None
+        _cost_map_source_info.loaded_at = datetime.now(timezone.utc)
+        adopt_model_cost_map(finalized)
+    except Exception as e:
+        verbose_logger.warning("LiteLLM: Background model cost map retry failed: %s", e)
+
+
 def get_model_cost_map(
     url: str,
     timeout: int = 5,
@@ -532,9 +601,7 @@ def get_model_cost_map(
     Public entry point — returns the model cost map dict.
 
     1. If ``LITELLM_LOCAL_MODEL_COST_MAP`` is set, uses the local backup only.
-    2. Otherwise fetches from ``url``, retrying transient HTTP errors
-       (429/5xx/transport) with Retry-After-aware backoff, validates
-       integrity, and falls back to the local backup on any failure.
+    2. Otherwise fetches from ``url``, retrying transient errors in a background thread.
 
     Only the backup model count is cached (a single int) for validation.
     The full backup dict is only parsed when it must be *returned* as a
@@ -553,24 +620,34 @@ def get_model_cost_map(
     _cost_map_source_info.url = url
     _cost_map_source_info.is_env_forced = False
 
-    result: Final = _fetch_remote_model_cost_map_with_retry_sync(
-        url=url,
-        timeout=timeout,
-        max_attempts=max_attempts,
-        sleep=sleep,
-        rng=rng if rng is not None else random.Random(),
-        client=client if client is not None else httpx,
-    )
-    if isinstance(result, ModelCostMapReloadUnavailable):
+    fetch_client: Final = client if client is not None else httpx
+    fetch_rng: Final = rng if rng is not None else random.Random()
+    outcome: Final = _attempt_fetch_sync(client=fetch_client, url=url, timeout=timeout)
+    if isinstance(outcome, _FetchAttemptRetryable) and max_attempts > 1:
+        threading.Thread(
+            target=_retry_remote_fetch_in_background,
+            kwargs={  # mutable-ok: threading requires a mutable keyword-arguments mapping
+                "url": url,
+                "timeout": timeout,
+                "max_attempts": max_attempts,
+                "sleep": sleep,
+                "rng": fetch_rng,
+                "client": fetch_client,
+                "first_outcome": outcome,
+            },
+            name="litellm-model-cost-map-retry",
+            daemon=True,
+        ).start()
+    if not isinstance(outcome, ModelCostMapReloaded):
         verbose_logger.warning(
             "LiteLLM: Failed to fetch remote model cost map from %s: %s. Falling back to local backup.",
             url,
-            result.reason,
+            outcome.reason,
         )
         _cost_map_source_info.source = "local"
-        _cost_map_source_info.fallback_reason = f"Remote fetch failed: {result.reason}"
+        _cost_map_source_info.fallback_reason = f"Remote fetch failed: {outcome.reason}"
         return _finalize_loaded_model_cost_map(GetModelCostMap.load_local_model_cost_map_with_revision()).model_cost_map
-    content: Final = result.model_cost_map
+    content: Final = outcome.model_cost_map
 
     # Validate using cached count (cheap int comparison, no file I/O)
     if not GetModelCostMap.validate_model_cost_map(
@@ -587,4 +664,4 @@ def get_model_cost_map(
 
     _cost_map_source_info.source = "remote"
     _cost_map_source_info.fallback_reason = None
-    return _finalize_loaded_model_cost_map(result).model_cost_map
+    return _finalize_loaded_model_cost_map(outcome).model_cost_map
diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index 7219b373dc3..7aed0553b4b 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -138,11 +138,7 @@ from litellm.types.utils import (
     TextCompletionResponse,
     TokenCountResponse,
 )
-from litellm.utils import (
-    _invalidate_model_cost_lowercase_map,
-    load_credentials_from_list,
-    reapply_runtime_model_cost_registrations,
-)
+from litellm.utils import load_credentials_from_list
 
 if TYPE_CHECKING:
     from aiohttp import ClientSession
@@ -4436,20 +4432,9 @@ def resolve_classifier_plugin(
 
 
 def _swap_in_model_cost_map(new_model_cost_map: dict) -> int:
-    """Adopt a freshly fetched cost map into this process's litellm state, return the model count"""
-    litellm.model_cost = new_model_cost_map
-    # Invalidate case-insensitive lookup map since model_cost was replaced
-    _invalidate_model_cost_lowercase_map()
-    # Repopulate provider model sets (e.g. litellm.anthropic_models) so that
-    # wildcard patterns like "anthropic/*" include any newly added models.
-    litellm.add_known_models(model_cost_map=new_model_cost_map)
-    # Counted before the re-apply below, which writes into this same dict, so the
-    # number reported describes the fetched price data alone.
-    fetched_model_count: Final = len(new_model_cost_map) if new_model_cost_map else 0
-    # The swap discards everything registered at runtime (deployment model_info,
-    # register_model overrides), so put it back on top of the fresh catalog.
-    reapply_runtime_model_cost_registrations()
-    return fetched_model_count
+    from litellm.litellm_core_utils.get_model_cost_map import adopt_model_cost_map
+
+    return adopt_model_cost_map(new_model_cost_map)
 
 
 def should_load_db_object(object_type: str | SupportedDBObjectType) -> bool:
diff --git a/litellm/utils.py b/litellm/utils.py
index 36b48d3b8d8..8df28870544 100644
--- a/litellm/utils.py
+++ b/litellm/utils.py
@@ -3079,7 +3079,7 @@ def register_model(
         # Convert stringified numbers to appropriate numeric types
         loaded_model_cost = model_cost
     elif isinstance(model_cost, str):
-        loaded_model_cost = litellm.get_model_cost_map(url=model_cost)
+        loaded_model_cost = litellm.get_model_cost_map(url=model_cost, max_attempts=1)
 
     if persist_across_reloads:
         _registrations: Final[Mapping[str, Mapping[str, object]]] = loaded_model_cost
diff --git a/tests/test_litellm/litellm_core_utils/test_get_model_cost_map.py b/tests/test_litellm/litellm_core_utils/test_get_model_cost_map.py
index 0495440c51c..266c2ca1465 100644
--- a/tests/test_litellm/litellm_core_utils/test_get_model_cost_map.py
+++ b/tests/test_litellm/litellm_core_utils/test_get_model_cost_map.py
@@ -6,6 +6,7 @@ count actual model entries, not reserved meta keys) and the extraction of the
 
 import json
 import os
+import threading
 
 import pytest
 
@@ -26,9 +27,7 @@ from litellm.litellm_core_utils.get_model_cost_map import (
 
 
 def _load_root_cost_map() -> dict:
-    path = os.path.join(
-        os.path.dirname(__file__), "../../../model_prices_and_context_window.json"
-    )
+    path = os.path.join(os.path.dirname(__file__), "../../../model_prices_and_context_window.json")
     with open(path) as f:
         return json.load(f)
 
@@ -44,9 +43,7 @@ def test_git_blob_id_is_what_git_hash_object_prints():
 
 
 def _make_models(n: int) -> dict:
-    return {
-        f"model-{i}": {"litellm_provider": "openai", "mode": "chat"} for i in range(n)
-    }
+    return {f"model-{i}": {"litellm_provider": "openai", "mode": "chat"} for i in range(n)}
 
 
 def test_count_model_entries_excludes_reserved_keys():
@@ -129,9 +126,7 @@ def test_finalize_pops_key_and_installs_rules():
 def test_finalize_with_no_block_clears_rules():
     previous = list(get_fallback_generalization_rules())
     try:
-        set_fallback_generalizations(
-            [{"name": "stale", "pattern": r"^x", "model_info": {"a": 1}}]
-        )
+        set_fallback_generalizations([{"name": "stale", "pattern": r"^x", "model_info": {"a": 1}}])
         _finalize_model_cost_map(_make_models(2))
         assert match_capability_generalizations("x-1") is None
     finally:
@@ -317,9 +312,7 @@ def test_get_model_cost_map_stamps_loaded_at():
 
     from litellm.litellm_core_utils import get_model_cost_map as module
 
-    client, _calls = _mock_client(
-        [httpx.Response(200, content=_real_map_bytes())], client_cls=httpx.Client
-    )
+    client, _calls = _mock_client([httpx.Response(200, content=_real_map_bytes())], client_cls=httpx.Client)
 
     before = datetime.now(timezone.utc)
     module.get_model_cost_map(url="https://example.invalid/cost_map.json", client=client)
@@ -328,6 +321,7 @@ def test_get_model_cost_map_stamps_loaded_at():
     assert loaded_at is not None
     assert before <= loaded_at <= datetime.now(timezone.utc)
 
+
 # ---------------------------------------------------------------------------
 # refetch_model_cost_map: retry/backoff behavior for runtime reloads
 # ---------------------------------------------------------------------------
@@ -394,9 +388,7 @@ async def test_refetch_retries_429_honoring_retry_after():
         ]
     )
     sleeper = _SleepRecorder()
-    result = await refetch_model_cost_map(
-        url=_URL, sleep=sleeper, rng=random.Random(0), client=client
-    )
+    result = await refetch_model_cost_map(url=_URL, sleep=sleeper, rng=random.Random(0), client=client)
     assert isinstance(result, ModelCostMapReloaded)
     assert len(result.model_cost_map) > 100
     assert calls["count"] == 3
@@ -408,9 +400,7 @@ async def test_refetch_gives_up_after_max_attempts_with_exponential_backoff():
     """All 429 without Retry-After: exponential backoff waits, then a failure value."""
     client, calls = _mock_client([httpx.Response(429)])
     sleeper = _SleepRecorder()
-    result = await refetch_model_cost_map(
-        url=_URL, sleep=sleeper, rng=random.Random(0), client=client
-    )
+    result = await refetch_model_cost_map(url=_URL, sleep=sleeper, rng=random.Random(0), client=client)
     assert isinstance(result, ModelCostMapReloadUnavailable)
     assert "429" in result.reason
     assert "after 3 attempts" in result.reason
@@ -430,9 +420,7 @@ async def test_refetch_caps_retry_after_wait():
         ]
     )
     sleeper = _SleepRecorder()
-    result = await refetch_model_cost_map(
-        url=_URL, sleep=sleeper, rng=random.Random(0), client=client
-    )
+    result = await refetch_model_cost_map(url=_URL, sleep=sleeper, rng=random.Random(0), client=client)
     assert isinstance(result, ModelCostMapReloaded)
     assert sleeper.waits == [30.0]
 
@@ -447,9 +435,7 @@ async def test_refetch_retries_transport_errors():
         ]
     )
     sleeper = _SleepRecorder()
-    result = await refetch_model_cost_map(
-        url=_URL, sleep=sleeper, rng=random.Random(0), client=client
-    )
+    result = await refetch_model_cost_map(url=_URL, sleep=sleeper, rng=random.Random(0), client=client)
     assert isinstance(result, ModelCostMapReloaded)
     assert calls["count"] == 2
     assert len(sleeper.waits) == 1
@@ -460,9 +446,7 @@ async def test_refetch_non_retryable_status_fails_immediately():
     """A 404 is permanent: one attempt, no sleeps, failure value."""
     client, calls = _mock_client([httpx.Response(404)])
     sleeper = _SleepRecorder()
-    result = await refetch_model_cost_map(
-        url=_URL, sleep=sleeper, rng=random.Random(0), client=client
-    )
+    result = await refetch_model_cost_map(url=_URL, sleep=sleeper, rng=random.Random(0), client=client)
     assert isinstance(result, ModelCostMapReloadUnavailable)
     assert "404" in result.reason
     assert calls["count"] == 1
@@ -473,9 +457,7 @@ async def test_refetch_non_retryable_status_fails_immediately():
 async def test_refetch_invalid_json_fails_immediately():
     client, calls = _mock_client([httpx.Response(200, content=b"not json")])
     sleeper = _SleepRecorder()
-    result = await refetch_model_cost_map(
-        url=_URL, sleep=sleeper, rng=random.Random(0), client=client
-    )
+    result = await refetch_model_cost_map(url=_URL, sleep=sleeper, rng=random.Random(0), client=client)
     assert isinstance(result, ModelCostMapReloadUnavailable)
     assert "invalid JSON" in result.reason
     assert calls["count"] == 1
@@ -487,9 +469,7 @@ async def test_refetch_shrunk_map_fails_integrity_not_swapped_in():
     """A drastically shrunk upstream file is rejected instead of being adopted."""
     tiny = json.dumps(_make_models(60)).encode()
     client, _calls = _mock_client([httpx.Response(200, content=tiny)])
-    result = await refetch_model_cost_map(
-        url=_URL, sleep=_SleepRecorder(), rng=random.Random(0), client=client
-    )
+    result = await refetch_model_cost_map(url=_URL, sleep=_SleepRecorder(), rng=random.Random(0), client=client)
     assert isinstance(result, ModelCostMapReloadUnavailable)
     assert "integrity validation" in result.reason
 
@@ -586,69 +566,125 @@ from litellm.litellm_core_utils.get_model_cost_map import (
 class _SyncSleepRecorder:
     """Injected in place of time.sleep so the boot path's waits are asserted without delay."""
 
-    def __init__(self):
+    def __init__(self, block=False):
         self.waits = []
+        self.block = block
+        self.started = threading.Event()
+        self.release = threading.Event()
 
     def __call__(self, seconds: float) -> None:
+        if self.block:
+            self.started.set()
+            self.release.wait(timeout=10)
         self.waits.append(seconds)
 
 
-def test_boot_load_retries_transient_failures_instead_of_falling_back():
-    """A refused connection then a 503 at pod boot used to pin the process to the bundled
-    backup for its lifetime; both are transient and must be retried before giving up."""
+def _retry_threads():
+    return [thread for thread in threading.enumerate() if thread.name == "litellm-model-cost-map-retry"]
+
+
+def test_boot_load_success_does_not_start_background_retry():
+    client, calls = _mock_client([httpx.Response(200, content=_real_map_bytes())], client_cls=httpx.Client)
+    sleeper = _SyncSleepRecorder()
+
+    cost_map = get_model_cost_map(
+        url=_URL,
+        sleep=sleeper,
+        rng=random.Random(0),
+        client=client,
+    )
+    assert calls["count"] == 1
+    assert sleeper.waits == []
+    assert _retry_threads() == []
+    assert cost_map.keys() >= _load_root_cost_map().keys() - {"sample_spec", FALLBACK_GENERALIZATIONS_KEY}
+    assert get_model_cost_map_source_info()["source"] == "remote"
+
+
+def test_boot_load_transient_failure_returns_local_then_background_retry_adopts_remote(monkeypatch):
+    import litellm
+    from litellm import utils as litellm_utils
+    from litellm.litellm_core_utils import get_model_cost_map as module
+
+    original_model_cost = litellm.model_cost
+    monkeypatch.setattr(litellm, "model_cost", dict(original_model_cost))
+    for name, provider_models in tuple(vars(litellm).items()):
+        if name.endswith("_models") and isinstance(provider_models, set):
+            monkeypatch.setattr(litellm, name, set(provider_models))
+    monkeypatch.setattr(litellm, "models_by_provider", dict(litellm.models_by_provider))
+    monkeypatch.setattr(
+        litellm_utils,
+        "_runtime_registered_model_cost",
+        dict(litellm_utils._runtime_registered_model_cost),
+    )
+    source_info = module._cost_map_source_info
+    for name in ("source", "url", "is_env_forced", "fallback_reason", "loaded_at", "source_revision", "etag"):
+        monkeypatch.setattr(source_info, name, getattr(source_info, name))
+
+    remote_map = _load_root_cost_map()
+    remote_map["claude-remote-only-test"] = {"litellm_provider": "anthropic", "mode": "chat"}
     client, calls = _mock_client(
         [
             httpx.ConnectError("connection refused"),
-            httpx.Response(503),
-            httpx.Response(200, content=_real_map_bytes()),
+            httpx.Response(200, content=json.dumps(remote_map).encode()),
         ],
         client_cls=httpx.Client,
     )
-    sleeper = _SyncSleepRecorder()
+    sleeper = _SyncSleepRecorder(block=True)
+    litellm.register_model({"my-runtime-model": {"litellm_provider": "custom", "max_input_tokens": 4321}})
 
-    cost_map = get_model_cost_map(url=_URL, sleep=sleeper, rng=random.Random(0), client=client)
-
-    assert calls["count"] == 3
-    assert len(sleeper.waits) == 2
-    assert 2.0 <= sleeper.waits[0] < 3.0
-    assert 4.0 <= sleeper.waits[1] < 5.0
-    source = get_model_cost_map_source_info()
-    assert source["source"] == "remote"
-    assert source["fallback_reason"] is None
-    assert cost_map.keys() >= _load_root_cost_map().keys() - {"sample_spec", FALLBACK_GENERALIZATIONS_KEY}
-
-
-def test_boot_load_honors_retry_after_then_falls_back_after_max_attempts():
-    """An outage longer than the retry budget still ends on the bundled backup, and the
-    recorded fallback reason says how many attempts were spent so operators can tell."""
-    client, calls = _mock_client(
-        [httpx.Response(429, headers={"Retry-After": "7"})], client_cls=httpx.Client
+    cost_map = get_model_cost_map(
+        url=_URL,
+        max_attempts=3,
+        sleep=sleeper,
+        rng=random.Random(0),
+        client=client,
     )
-    sleeper = _SyncSleepRecorder()
 
-    cost_map = get_model_cost_map(url=_URL, sleep=sleeper, rng=random.Random(0), client=client)
-
-    assert calls["count"] == 3
-    assert sleeper.waits == [7.0, 7.0]
-    source = get_model_cost_map_source_info()
-    assert source["source"] == "local"
-    assert "after 3 attempts" in source["fallback_reason"]
-    assert len(cost_map) > 100
+    assert calls["count"] == 1
+    assert sleeper.waits == []
+    assert sleeper.started.wait(timeout=10)
+    threads = _retry_threads()
+    try:
+        assert len(threads) == 1
+        assert "claude-remote-only-test" not in cost_map
+        assert cost_map.keys() == _finalize_model_cost_map(GetModelCostMap.load_local_model_cost_map()).keys()
+        source = get_model_cost_map_source_info()
+        assert source["source"] == "local"
+        assert source["fallback_reason"].startswith("Remote fetch failed:")
+        sleeper.release.set()
+        for thread in threads:
+            thread.join(timeout=10)
+        assert all(not thread.is_alive() for thread in threads)
+        assert sleeper.waits and 2.0 <= sleeper.waits[0] < 3.0
+        assert calls["count"] == 2
+        assert "claude-remote-only-test" in litellm.model_cost
+        assert "claude-remote-only-test" in litellm.anthropic_models
+        assert "my-runtime-model" in litellm.model_cost
+        source = get_model_cost_map_source_info()
+        assert source["source"] == "remote"
+        assert source["fallback_reason"] is None
+    finally:
+        sleeper.release.set()
+        for thread in _retry_threads():
+            thread.join(timeout=10)
 
 
-def test_boot_load_does_not_retry_permanent_failures():
-    """A 404 or a malformed URL cannot heal by waiting: one attempt, no sleeps, backup."""
+def test_boot_load_does_not_retry_non_retryable_failure():
     client, calls = _mock_client([httpx.Response(404)], client_cls=httpx.Client)
     sleeper = _SyncSleepRecorder()
 
-    get_model_cost_map(url=_URL, sleep=sleeper, rng=random.Random(0), client=client)
+    get_model_cost_map(
+        url=_URL,
+        sleep=sleeper,
+        rng=random.Random(0),
+        client=client,
+    )
     assert calls["count"] == 1
     assert sleeper.waits == []
-    assert get_model_cost_map_source_info()["source"] == "local"
-
-    get_model_cost_map(url="not a url", sleep=sleeper, rng=random.Random(0))
-    assert sleeper.waits == []
-    assert get_model_cost_map_source_info()["source"] == "local"
+    assert _retry_threads() == []
+    source = get_model_cost_map_source_info()
+    assert source["source"] == "local"
+    assert source["fallback_reason"] is not None
 
 
 def test_boot_load_respects_local_env_override(monkeypatch):
@@ -701,7 +737,9 @@ def test_boot_load_that_fails_the_integrity_check_reports_the_backup_not_the_rej
     )
     get_model_cost_map(url=_URL, sleep=_SyncSleepRecorder(), rng=random.Random(0), client=remote)
     shrunk_body = b'{"gpt-5.4-mini": {"mode": "chat", "input_cost_per_token": 1e-06, "output_cost_per_token": 2e-06}}'
-    shrunk, _ = _mock_client([httpx.Response(200, headers={"ETag": 'W/"shrunk"'}, content=shrunk_body)], client_cls=httpx.Client)
+    shrunk, _ = _mock_client(
+        [httpx.Response(200, headers={"ETag": 'W/"shrunk"'}, content=shrunk_body)], client_cls=httpx.Client
+    )
 
     get_model_cost_map(url=_URL, sleep=_SyncSleepRecorder(), rng=random.Random(0), client=shrunk)
 
diff --git a/tests/test_litellm/test_utils.py b/tests/test_litellm/test_utils.py
index fee5e3a2e4c..e42608c9904 100644
--- a/tests/test_litellm/test_utils.py
+++ b/tests/test_litellm/test_utils.py
@@ -2,10 +2,12 @@ import asyncio
 import json
 import logging
 import os
+import threading
 from datetime import datetime, timedelta, timezone
 from typing import Final
 from unittest.mock import AsyncMock, MagicMock, patch
 
+import httpx
 import pytest
 import respx
 from jsonschema import validate
@@ -2382,6 +2384,28 @@ def test_register_model_with_scientific_notation():
     _invalidate_model_cost_lowercase_map()
 
 
+@respx.mock
+def test_register_model_url_fetch_uses_single_attempt(monkeypatch):
+    monkeypatch.delenv("LITELLM_LOCAL_MODEL_COST_MAP", raising=False)
+    monkeypatch.setattr(litellm, "model_cost", dict(litellm.model_cost))
+    before = dict(litellm.model_cost)
+    threads_before = {thread.name for thread in threading.enumerate()}
+    route = respx.get("https://example.invalid/custom_pricing.json").mock(
+        return_value=httpx.Response(503)
+    )
+
+    litellm.register_model(model_cost="https://example.invalid/custom_pricing.json")
+
+    threads_after = {thread.name for thread in threading.enumerate()}
+    assert route.call_count == 1
+    assert not (threads_after - threads_before) & {"litellm-model-cost-map-retry"}
+    assert not any(
+        thread.name == "litellm-model-cost-map-retry" and thread.is_alive()
+        for thread in threading.enumerate()
+    )
+    assert litellm.model_cost.keys() >= before.keys()
+
+
 def test_register_model_openrouter_without_slash():
     """
     Test that register_model handles openrouter models without '/' in the name.