diff --git a/litellm/cost_calculator.py b/litellm/cost_calculator.py
index d31df2a82a5..12ad598f0e7 100644
--- a/litellm/cost_calculator.py
+++ b/litellm/cost_calculator.py
@@ -1905,7 +1905,8 @@ def _image_cost_from_token_usage(
 ) -> Optional[float]:
     """Token-based image cost from ``cost_info`` + ``image_response.usage``.
 
-    Returns ``None`` when no token-cost keys match any non-zero token count.
+    Returns ``None`` only when ``cost_info`` declares no recognised per-token
+    rate keys; zero-cost models that explicitly declare zero rates return ``0.0``.
     """
     usage = getattr(image_response, "usage", None)
     if usage is None:
@@ -1923,16 +1924,23 @@ def _image_cost_from_token_usage(
     cached_in = _detail("prompt_tokens_details", "cached_tokens")
     image_in = _detail("prompt_tokens_details", "image_tokens")
     image_out = _detail("completion_tokens_details", "image_tokens")
-    text_in_uncached = max(text_in - cached_in, 0)
+
+    if image_in > 0 and cost_info.get("cache_read_input_image_token_cost") is not None:
+        text_in_uncached = text_in
+        cache_rate_key = "cache_read_input_image_token_cost"
+    else:
+        text_in_uncached = max(text_in - cached_in, 0)
+        cache_rate_key = "cache_read_input_token_cost"
 
     rates: List[Tuple[str, int]] = [
         ("input_cost_per_token", text_in_uncached),
-        ("cache_read_input_token_cost", cached_in),
+        (cache_rate_key, cached_in),
         ("input_cost_per_image_token", image_in),
         ("output_cost_per_image_token", image_out),
     ]
-    cost = sum((cost_info.get(key) or 0) * tokens for key, tokens in rates)
-    return cost if cost > 0 else None
+    if not any(cost_info.get(key) is not None for key, _ in rates):
+        return None
+    return sum((cost_info.get(key) or 0) * tokens for key, tokens in rates)
 
 
 def default_image_cost_calculator(
diff --git a/tests/test_litellm/test_default_image_cost_calculator.py b/tests/test_litellm/test_default_image_cost_calculator.py
index 1cc5c0eb328..68a930533f1 100644
--- a/tests/test_litellm/test_default_image_cost_calculator.py
+++ b/tests/test_litellm/test_default_image_cost_calculator.py
@@ -206,6 +206,65 @@ class TestDefaultImageCostCalculator:
         expected = 40 * 5e-6 + 160 * 1.25e-6 + 600 * 3e-5
         assert abs(cost - expected) < 1e-9
 
+    def test_token_fallback_uses_image_cache_rate_when_declared(self, monkeypatch):
+        """Image-edit responses report a single ``cached_tokens`` count;
+        when the model declares ``cache_read_input_image_token_cost`` and
+        image input tokens are present, that rate is used (and text input
+        is billed in full).
+        """
+        monkeypatch.setitem(
+            litellm.model_cost,
+            "synthetic-image-cache-model",
+            {
+                "input_cost_per_token": 5e-6,
+                "cache_read_input_token_cost": 1.25e-6,
+                "input_cost_per_image_token": 8e-6,
+                "cache_read_input_image_token_cost": 2e-6,
+                "output_cost_per_image_token": 3e-5,
+                "litellm_provider": "openai",
+                "mode": "image_generation",
+            },
+        )
+
+        cost = default_image_cost_calculator(
+            model="openai/synthetic-image-cache-model",
+            custom_llm_provider="openai",
+            quality="high",
+            n=1,
+            size="1280x720",
+            image_response=_image_response(
+                text_in=510, image_in=1452, cached_in=300, image_out=5488
+            ),
+        )
+        # text fully uncached, cached_tokens charged at image cache rate
+        expected = 510 * 5e-6 + 300 * 2e-6 + 1452 * 8e-6 + 5488 * 3e-5
+        assert abs(cost - expected) < 1e-9
+
+    def test_token_fallback_returns_zero_for_free_model(self, monkeypatch):
+        """A model that declares zero per-token rates is genuinely free —
+        the fallback must return ``0.0`` rather than raising.
+        """
+        monkeypatch.setitem(
+            litellm.model_cost,
+            "synthetic-free-image-model",
+            {
+                "input_cost_per_token": 0.0,
+                "output_cost_per_image_token": 0.0,
+                "litellm_provider": "openai",
+                "mode": "image_generation",
+            },
+        )
+
+        cost = default_image_cost_calculator(
+            model="openai/synthetic-free-image-model",
+            custom_llm_provider="openai",
+            quality="low",
+            n=1,
+            size="2048x768",
+            image_response=_image_response(text_in=25, image_out=772),
+        )
+        assert cost == 0.0
+
     def test_unmapped_model_without_image_response_raises(self):
         """Negative: cost map miss + no ``image_response`` to fall back on
         — preserve the original behaviour of raising rather than silently