Merge a0bfdc85d3 into 8d6b8dab3d

2026-10-29 06:33:28 +00:00 · 2026-10-21 03:37:33 +00:00 · 2026-10-21 03:37:33 +00:00 · 7b93c33a1b
commit 7b93c33a1b
parent 8d6b8dab3d a0bfdc85d3
6 changed files with 192 additions and 15 deletions
--- a/litellm/cost_calculator.py
+++ b/litellm/cost_calculator.py
@ -1787,6 +1787,7 @@ def completion_cost(
                        call_type=call_type,
                        model_info=_deployment_model_info(litellm_logging_obj, custom_pricing, router_model_id),
                        vertex_location=vertex_location,
+                        data_residency=data_residency,
                    )
                elif call_type in _VIDEO_CALL_TYPES:
                    ### VIDEO GENERATION COST CALCULATION ###
--- a/litellm/litellm_core_utils/llm_cost_calc/utils.py
+++ b/litellm/litellm_core_utils/llm_cost_calc/utils.py
@ -1933,6 +1933,7 @@ def calculate_image_response_cost_from_usage(
    custom_llm_provider: str,
    model_info: ModelInfo | None = None,
    vertex_location: str | None = None,
+    data_residency: str | None = None,
 ) -> float | None:
    """
    Calculate image generation cost from usage metadata when available.
@ -2017,6 +2018,7 @@ def calculate_image_response_cost_from_usage(
        custom_llm_provider=custom_llm_provider,
        model_info=model_info,
        vertex_location=vertex_location,
+        data_residency=data_residency,
    )
    return prompt_cost + completion_cost

@ -2081,6 +2083,7 @@ class CostCalculatorUtils:
        call_type: str | None = None,
        model_info: ModelInfo | None = None,
        vertex_location: str | None = None,
+        data_residency: str | None = None,
    ) -> float:
        """
        Route the image generation cost calculator based on the custom_llm_provider
@ -2232,6 +2235,7 @@ class CostCalculatorUtils:
                    image_response=completion_response,
                    custom_llm_provider=custom_llm_provider,
                    model_info=pricing,
+                    data_residency=data_residency,
                )
            # Fall through to default for DALL-E models
            return default_image_cost_calculator(
--- a/litellm/llms/openai/image_generation/cost_calculator.py
+++ b/litellm/llms/openai/image_generation/cost_calculator.py
@ -21,6 +21,7 @@ def cost_calculator(
    image_response: ImageResponse,
    custom_llm_provider: str | None = None,
    model_info: ModelInfo | None = None,
+    data_residency: str | None = None,
 ) -> float:
    """Calculate cost for OpenAI gpt-image models (token-based pricing)."""
    provider: Final = custom_llm_provider or "openai"
@ -38,7 +39,11 @@ def cost_calculator(
    # A chat Usage with an explicit output breakdown: cost via generic_cost_per_token.
    if isinstance(usage, Usage) and usage.completion_tokens_details is not None:
        prompt_cost, completion_cost = generic_cost_per_token(
-            model=model, usage=usage, custom_llm_provider=provider, model_info=price_table
+            model=model,
+            usage=usage,
+            custom_llm_provider=provider,
+            model_info=price_table,
+            data_residency=data_residency,
        )
        return prompt_cost + completion_cost

@ -48,7 +53,11 @@ def cost_calculator(
    # does not itemize output and splitting text/image when it does.
    if getattr(usage, "input_tokens", None) is not None:
        token_based_cost: Final = calculate_image_response_cost_from_usage(
-            model=model, image_response=image_response, custom_llm_provider=provider, model_info=price_table
+            model=model,
+            image_response=image_response,
+            custom_llm_provider=provider,
+            model_info=price_table,
+            data_residency=data_residency,
        )
        if token_based_cost is not None:
            return token_based_cost
@ -57,7 +66,11 @@ def cost_calculator(
    # cost via generic_cost_per_token (text rate) instead of returning 0.0.
    if isinstance(usage, Usage):
        prompt_cost, completion_cost = generic_cost_per_token(
-            model=model, usage=usage, custom_llm_provider=provider, model_info=price_table
+            model=model,
+            usage=usage,
+            custom_llm_provider=provider,
+            model_info=price_table,
+            data_residency=data_residency,
        )
        return prompt_cost + completion_cost

--- a/litellm/model_prices_and_context_window_backup.json
+++ b/litellm/model_prices_and_context_window_backup.json
@ -35678,7 +35678,9 @@
            "/v1/images/edits"
        ],
        "supports_vision": true,
-        "supports_pdf_input": true
+        "supports_pdf_input": true,
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "gpt-image-2-2026-04-21": {
        "cache_read_input_image_token_cost": 2e-06,
@ -35694,7 +35696,9 @@
            "/v1/images/edits"
        ],
        "supports_vision": true,
-        "supports_pdf_input": true
+        "supports_pdf_input": true,
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "gpt-image-2.5-flare": {
        "cache_read_input_image_token_cost": 2e-06,
@ -35713,7 +35717,9 @@
        ],
        "supports_vision": true,
        "supports_pdf_input": true,
-        "source": "https://developers.openai.com/api/docs/pricing"
+        "source": "https://developers.openai.com/api/docs/pricing",
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "gpt-image-2.5-flare-2026-09-08": {
        "cache_read_input_image_token_cost": 2e-06,
@ -35729,7 +35735,9 @@
        ],
        "supports_vision": true,
        "supports_pdf_input": true,
-        "source": "https://developers.openai.com/api/docs/pricing"
+        "source": "https://developers.openai.com/api/docs/pricing",
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "gpt-image-2.5-sunburst": {
        "cache_read_input_image_token_cost": 2e-06,
@ -35748,7 +35756,9 @@
        ],
        "supports_vision": true,
        "supports_pdf_input": true,
-        "source": "https://developers.openai.com/api/docs/pricing"
+        "source": "https://developers.openai.com/api/docs/pricing",
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "gpt-image-2.5-sunburst-2026-09-08": {
        "cache_read_input_image_token_cost": 2e-06,
@ -35764,7 +35774,9 @@
        ],
        "supports_vision": true,
        "supports_pdf_input": true,
-        "source": "https://developers.openai.com/api/docs/pricing"
+        "source": "https://developers.openai.com/api/docs/pricing",
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "low/1024-x-1024/gpt-image-1.5": {
        "deprecation_date": "2026-12-01",
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@ -35678,7 +35678,9 @@
            "/v1/images/edits"
        ],
        "supports_vision": true,
-        "supports_pdf_input": true
+        "supports_pdf_input": true,
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "gpt-image-2-2026-04-21": {
        "cache_read_input_image_token_cost": 2e-06,
@ -35694,7 +35696,9 @@
            "/v1/images/edits"
        ],
        "supports_vision": true,
-        "supports_pdf_input": true
+        "supports_pdf_input": true,
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "gpt-image-2.5-flare": {
        "cache_read_input_image_token_cost": 2e-06,
@ -35713,7 +35717,9 @@
        ],
        "supports_vision": true,
        "supports_pdf_input": true,
-        "source": "https://developers.openai.com/api/docs/pricing"
+        "source": "https://developers.openai.com/api/docs/pricing",
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "gpt-image-2.5-flare-2026-09-08": {
        "cache_read_input_image_token_cost": 2e-06,
@ -35729,7 +35735,9 @@
        ],
        "supports_vision": true,
        "supports_pdf_input": true,
-        "source": "https://developers.openai.com/api/docs/pricing"
+        "source": "https://developers.openai.com/api/docs/pricing",
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "gpt-image-2.5-sunburst": {
        "cache_read_input_image_token_cost": 2e-06,
@ -35748,7 +35756,9 @@
        ],
        "supports_vision": true,
        "supports_pdf_input": true,
-        "source": "https://developers.openai.com/api/docs/pricing"
+        "source": "https://developers.openai.com/api/docs/pricing",
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "gpt-image-2.5-sunburst-2026-09-08": {
        "cache_read_input_image_token_cost": 2e-06,
@ -35764,7 +35774,9 @@
        ],
        "supports_vision": true,
        "supports_pdf_input": true,
-        "source": "https://developers.openai.com/api/docs/pricing"
+        "source": "https://developers.openai.com/api/docs/pricing",
+        "regional_processing_uplift_multiplier_eu": 1.1,
+        "regional_processing_uplift_multiplier_us": 1.1
    },
    "low/1024-x-1024/gpt-image-1.5": {
        "deprecation_date": "2026-12-01",
--- a/tests/unit/test_image_cost_data_residency.py
+++ b/tests/unit/test_image_cost_data_residency.py
@ -0,0 +1,135 @@
+"""
+Tests for the OpenAI regional-processing uplift on gpt-image spend
+
+This tests the fix for GitHub issue #46407:
+https://github.com/BerriAI/litellm/issues/46407
+
+`data_residency: eu` (routed through eu.api.openai.com) must apply the
+per-model regional uplift multiplier to image-generation spend exactly as it
+already does for chat/responses spend.
+"""
+
+import pytest
+
+import litellm
+from litellm.types.utils import (
+    ImageObject,
+    ImageResponse,
+    ImageUsage,
+    ImageUsageInputTokensDetails,
+)
+
+
+@pytest.fixture(autouse=True)
+def _use_local_model_cost_map(monkeypatch):
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    monkeypatch.setattr(
+        litellm, "model_cost", litellm.get_model_cost_map(url="")
+    )
+    litellm.get_model_info.cache_clear()
+    yield
+    litellm.get_model_info.cache_clear()
+
+
+def _usage() -> ImageUsage:
+    return ImageUsage(
+        input_tokens=100,
+        input_tokens_details=ImageUsageInputTokensDetails(image_tokens=50, text_tokens=50),
+        output_tokens=200,
+        total_tokens=300,
+    )
+
+
+def _image_response() -> ImageResponse:
+    return ImageResponse(
+        created=1234567890,
+        data=[ImageObject(url="http://example.com/image.jpg")],
+        usage=_usage(),
+    )
+
+
+class TestImageDataResidencyUplift:
+    def test_image_usage_path_applies_eu_uplift(self) -> None:
+        """The ImageUsage path (calculate_image_response_cost_from_usage) must apply the EU multiplier."""
+        from litellm.llms.openai.image_generation.cost_calculator import cost_calculator
+
+        image_response = _image_response()
+        base_cost = cost_calculator(
+            model="gpt-image-2.5-flare",
+            image_response=image_response,
+            custom_llm_provider="openai",
+        )
+        eu_cost = cost_calculator(
+            model="gpt-image-2.5-flare",
+            image_response=image_response,
+            custom_llm_provider="openai",
+            data_residency="eu",
+        )
+
+        assert base_cost > 0
+        assert eu_cost == pytest.approx(base_cost * 1.1)
+
+    def test_route_from_completion_cost_applies_eu_uplift(self) -> None:
+        """completion_cost forwards data_residency into the image branch (issue #46407 part 1)."""
+        base_cost = litellm.completion_cost(
+            model="gpt-image-2.5-flare",
+            custom_llm_provider="openai",
+            completion_response=_image_response(),
+            call_type="aimage_generation",
+        )
+        eu_cost = litellm.completion_cost(
+            model="gpt-image-2.5-flare",
+            custom_llm_provider="openai",
+            completion_response=_image_response(),
+            call_type="aimage_generation",
+            data_residency="eu",
+        )
+
+        assert base_cost > 0
+        assert eu_cost == pytest.approx(base_cost * 1.1)
+
+    def test_us_uplift_applies_and_unknown_residency_does_not(self) -> None:
+        from litellm.llms.openai.image_generation.cost_calculator import cost_calculator
+
+        image_response = _image_response()
+        base_cost = cost_calculator(
+            model="gpt-image-2.5-flare",
+            image_response=image_response,
+            custom_llm_provider="openai",
+        )
+
+        us_cost = cost_calculator(
+            model="gpt-image-2.5-flare",
+            image_response=image_response,
+            custom_llm_provider="openai",
+            data_residency="us",
+        )
+        unknown_cost = cost_calculator(
+            model="gpt-image-2.5-flare",
+            image_response=image_response,
+            custom_llm_provider="openai",
+            data_residency="ap-southeast-1",
+        )
+
+        assert us_cost == pytest.approx(base_cost * 1.1)
+        assert unknown_cost == pytest.approx(base_cost)
+
+    def test_pre_cutoff_gpt_image_model_has_no_uplift(self) -> None:
+        """gpt-image-1.5 predates the 2026-03-05 cutoff and carries no multiplier: residency must not change spend."""
+        from litellm.llms.openai.image_generation.cost_calculator import cost_calculator
+
+        image_response = _image_response()
+        base_cost = cost_calculator(
+            model="gpt-image-1.5",
+            image_response=image_response,
+            custom_llm_provider="openai",
+        )
+        eu_cost = cost_calculator(
+            model="gpt-image-1.5",
+            image_response=image_response,
+            custom_llm_provider="openai",
+            data_residency="eu",
+        )
+
+        assert base_cost > 0
+        assert eu_cost == pytest.approx(base_cost)