fix(cost): add missing cache_read_input_token_cost for DeepSeek models

Several DeepSeek and OpenRouter/DeepSeek models had cache pricing data
under input_cost_per_token_cache_hit but not under the key the cost
calculator actually uses (cache_read_input_token_cost). This caused
cost_breakdown to omit cache_read_cost for these models.

Also adds cache_creation_input_token_cost=0.0 for all DeepSeek provider
models since DeepSeek auto-caches with no creation cost.

Closes #31594

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/model_prices_and_context_window_backup.json

@@ -11255,6 +11255,7 @@
         ]
     },
     "deepseek-chat": {
+        "cache_creation_input_token_cost": 0.0,
         "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
         "litellm_provider": "deepseek",
@@ -11276,6 +11277,7 @@
         "supports_tool_choice": true
     },
     "deepseek-reasoner": {
+        "cache_creation_input_token_cost": 0.0,
         "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
         "litellm_provider": "deepseek",
@@ -13806,6 +13808,8 @@
         "supports_tool_choice": true
     },
     "deepseek/deepseek-coder": {
+        "cache_creation_input_token_cost": 0.0,
+        "cache_read_input_token_cost": 1.4e-08,
         "input_cost_per_token": 1.4e-07,
         "input_cost_per_token_cache_hit": 1.4e-08,
         "litellm_provider": "deepseek",
@@ -13820,6 +13824,8 @@
         "supports_tool_choice": true
     },
     "deepseek/deepseek-r1": {
+        "cache_creation_input_token_cost": 0.0,
+        "cache_read_input_token_cost": 1.4e-07,
         "input_cost_per_token": 5.5e-07,
         "input_cost_per_token_cache_hit": 1.4e-07,
         "litellm_provider": "deepseek",
@@ -13835,6 +13841,7 @@
         "supports_tool_choice": true
     },
     "deepseek/deepseek-reasoner": {
+        "cache_creation_input_token_cost": 0.0,
         "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
         "input_cost_per_token_cache_hit": 2.8e-08,
@@ -13875,6 +13882,8 @@
         "supports_tool_choice": true
     },
     "deepseek/deepseek-v3.2": {
+        "cache_creation_input_token_cost": 0.0,
+        "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
         "input_cost_per_token_cache_hit": 2.8e-08,
         "litellm_provider": "deepseek",
@@ -28324,6 +28333,7 @@
         "supports_tool_choice": true
     },
     "openrouter/deepseek/deepseek-chat-v3.1": {
+        "cache_read_input_token_cost": 2e-08,
         "input_cost_per_token": 2e-07,
         "input_cost_per_token_cache_hit": 2e-08,
         "litellm_provider": "openrouter",
@@ -28339,6 +28349,7 @@
         "supports_tool_choice": true
     },
     "openrouter/deepseek/deepseek-v3.2": {
+        "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
         "input_cost_per_token_cache_hit": 2.8e-08,
         "litellm_provider": "openrouter",
@@ -28354,6 +28365,7 @@
         "supports_tool_choice": true
     },
     "openrouter/deepseek/deepseek-v3.2-exp": {
+        "cache_read_input_token_cost": 2e-08,
         "input_cost_per_token": 2e-07,
         "input_cost_per_token_cache_hit": 2e-08,
         "litellm_provider": "openrouter",
@@ -28369,6 +28381,7 @@
         "supports_tool_choice": true
     },
     "openrouter/deepseek/deepseek-r1": {
+        "cache_read_input_token_cost": 1.4e-07,
         "input_cost_per_token": 5.5e-07,
         "input_cost_per_token_cache_hit": 1.4e-07,
         "litellm_provider": "openrouter",
@@ -28384,6 +28397,7 @@
         "supports_tool_choice": true
     },
     "openrouter/deepseek/deepseek-r1-0528": {
+        "cache_read_input_token_cost": 1.4e-07,
         "input_cost_per_token": 5e-07,
         "input_cost_per_token_cache_hit": 1.4e-07,
         "litellm_provider": "openrouter",

model_prices_and_context_window.json

@@ -11255,6 +11255,7 @@
         ]
     },
     "deepseek-chat": {
+        "cache_creation_input_token_cost": 0.0,
         "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
         "litellm_provider": "deepseek",
@@ -11276,6 +11277,7 @@
         "supports_tool_choice": true
     },
     "deepseek-reasoner": {
+        "cache_creation_input_token_cost": 0.0,
         "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
         "litellm_provider": "deepseek",
@@ -13806,6 +13808,8 @@
         "supports_tool_choice": true
     },
     "deepseek/deepseek-coder": {
+        "cache_creation_input_token_cost": 0.0,
+        "cache_read_input_token_cost": 1.4e-08,
         "input_cost_per_token": 1.4e-07,
         "input_cost_per_token_cache_hit": 1.4e-08,
         "litellm_provider": "deepseek",
@@ -13820,6 +13824,8 @@
         "supports_tool_choice": true
     },
     "deepseek/deepseek-r1": {
+        "cache_creation_input_token_cost": 0.0,
+        "cache_read_input_token_cost": 1.4e-07,
         "input_cost_per_token": 5.5e-07,
         "input_cost_per_token_cache_hit": 1.4e-07,
         "litellm_provider": "deepseek",
@@ -13835,6 +13841,7 @@
         "supports_tool_choice": true
     },
     "deepseek/deepseek-reasoner": {
+        "cache_creation_input_token_cost": 0.0,
         "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
         "input_cost_per_token_cache_hit": 2.8e-08,
@@ -13875,6 +13882,8 @@
         "supports_tool_choice": true
     },
     "deepseek/deepseek-v3.2": {
+        "cache_creation_input_token_cost": 0.0,
+        "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
         "input_cost_per_token_cache_hit": 2.8e-08,
         "litellm_provider": "deepseek",
@@ -28485,6 +28494,7 @@
         "supports_tool_choice": true
     },
     "openrouter/deepseek/deepseek-chat-v3.1": {
+        "cache_read_input_token_cost": 2e-08,
         "input_cost_per_token": 2e-07,
         "input_cost_per_token_cache_hit": 2e-08,
         "litellm_provider": "openrouter",
@@ -28500,6 +28510,7 @@
         "supports_tool_choice": true
     },
     "openrouter/deepseek/deepseek-v3.2": {
+        "cache_read_input_token_cost": 2.8e-08,
         "input_cost_per_token": 2.8e-07,
         "input_cost_per_token_cache_hit": 2.8e-08,
         "litellm_provider": "openrouter",
@@ -28515,6 +28526,7 @@
         "supports_tool_choice": true
     },
     "openrouter/deepseek/deepseek-v3.2-exp": {
+        "cache_read_input_token_cost": 2e-08,
         "input_cost_per_token": 2e-07,
         "input_cost_per_token_cache_hit": 2e-08,
         "litellm_provider": "openrouter",
@@ -28530,6 +28542,7 @@
         "supports_tool_choice": true
     },
     "openrouter/deepseek/deepseek-r1": {
+        "cache_read_input_token_cost": 1.4e-07,
         "input_cost_per_token": 5.5e-07,
         "input_cost_per_token_cache_hit": 1.4e-07,
         "litellm_provider": "openrouter",
@@ -28545,6 +28558,7 @@
         "supports_tool_choice": true
     },
     "openrouter/deepseek/deepseek-r1-0528": {
+        "cache_read_input_token_cost": 1.4e-07,
         "input_cost_per_token": 5e-07,
         "input_cost_per_token_cache_hit": 1.4e-07,
         "litellm_provider": "openrouter",

tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py

@@ -2038,3 +2038,43 @@ def test_token_type_cost_breakdown_applies_regional_uplift():
     text_input_cost = 600 * model_info["input_cost_per_token"] * uplift
     assert text_output_cost + eu.reasoning_cost == pytest.approx(completion_cost)
     assert text_input_cost + eu.cache_read_cost == pytest.approx(prompt_cost)
+
+
+@pytest.mark.parametrize(
+    "model,custom_llm_provider,expected_cache_read_rate",
+    [
+        ("deepseek-chat", "deepseek", 2.8e-08),
+        ("deepseek/deepseek-r1", "deepseek", 1.4e-07),
+        ("deepseek/deepseek-v3.2", "deepseek", 2.8e-08),
+        ("deepseek/deepseek-coder", "deepseek", 1.4e-08),
+    ],
+)
+def test_deepseek_cache_read_cost_in_breakdown(
+    model, custom_llm_provider, expected_cache_read_rate
+):
+    """
+    DeepSeek models report cached tokens via prompt_cache_hit_tokens. The
+    cost breakdown must produce a non-zero cache_read_cost matching the
+    model's cache_read_input_token_cost rate.
+
+    Regression for https://github.com/BerriAI/litellm/issues/31594
+    """
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
+
+    cache_hit_tokens = 64
+    usage = Usage(
+        prompt_tokens=100,
+        completion_tokens=50,
+        total_tokens=150,
+        prompt_cache_hit_tokens=cache_hit_tokens,
+        prompt_cache_miss_tokens=36,
+    )
+
+    breakdown = get_token_type_cost_breakdown(
+        model=model, custom_llm_provider=custom_llm_provider, usage=usage
+    )
+
+    assert breakdown.cache_read_cost == pytest.approx(
+        cache_hit_tokens * expected_cache_read_rate
+    )

tests/test_litellm/test_cost_calculator.py

@@ -3176,3 +3176,79 @@ def test_completion_cost_logs_reasoning_and_cache_breakdown():
     assert logging_obj.cost_breakdown is not None
     assert logging_obj.cost_breakdown["reasoning_cost"] == pytest.approx(3114 * 2.5e-06)
     assert logging_obj.cost_breakdown["cache_read_cost"] == pytest.approx(100 * 3e-08)
+
+
+@pytest.mark.parametrize(
+    "model,custom_llm_provider,cache_read_rate",
+    [
+        ("deepseek-chat", "deepseek", 2.8e-08),
+        ("deepseek/deepseek-r1", "deepseek", 1.4e-07),
+        ("deepseek/deepseek-v3.2", "deepseek", 2.8e-08),
+        ("deepseek/deepseek-coder", "deepseek", 1.4e-08),
+    ],
+)
+def test_deepseek_cost_breakdown_includes_cache_read_cost(
+    model, custom_llm_provider, cache_read_rate
+):
+    """
+    DeepSeek reports cached tokens via prompt_cache_hit_tokens. The cost
+    breakdown must surface cache_read_cost so spend logs show where cache
+    savings come from, not just a lower total.
+
+    Regression for https://github.com/BerriAI/litellm/issues/31594
+    """
+    from datetime import datetime
+
+    from litellm.litellm_core_utils.litellm_logging import Logging
+    from litellm.types.utils import Choices, Message
+
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
+
+    cache_hit_tokens = 64
+
+    logging_obj = Logging(
+        model=model,
+        messages=[{"role": "user", "content": "Hello"}],
+        stream=False,
+        call_type="completion",
+        start_time=datetime.now(),
+        litellm_call_id="deepseek-cache-breakdown",
+        function_id="f",
+    )
+
+    response = ModelResponse(
+        id="x",
+        created=1,
+        model=model,
+        object="chat.completion",
+        choices=[
+            Choices(
+                index=0,
+                message=Message(role="assistant", content="hi"),
+                finish_reason="stop",
+            )
+        ],
+        usage=Usage(
+            prompt_tokens=100,
+            completion_tokens=50,
+            total_tokens=150,
+            prompt_cache_hit_tokens=cache_hit_tokens,
+            prompt_cache_miss_tokens=36,
+        ),
+    )
+
+    litellm.completion_cost(
+        completion_response=response,
+        model=model,
+        custom_llm_provider=custom_llm_provider,
+        litellm_logging_obj=logging_obj,
+    )
+
+    assert logging_obj.cost_breakdown is not None
+    assert "cache_read_cost" in logging_obj.cost_breakdown, (
+        f"cache_read_cost missing from cost_breakdown for {model}"
+    )
+    assert logging_obj.cost_breakdown["cache_read_cost"] == pytest.approx(
+        cache_hit_tokens * cache_read_rate
+    )